Transcript[TEASER][MUSIC PLAYS UNDER DIALOGUE]AKSHAY NAMBI: For me, research is just not about pushing the boundaries of the knowledge. Its about ensuring that these advancements translate to meaningful impact on the ground. So, yes, the big goals that guide most of my work is twofold. One, how do we build technology thats scaled to benefit large populations? And two, at the same time, Im motivated by the challenge of tackling complex problems. That provides opportunity to explore, learn, and also create something new, and thats what keeps me excited.[TEASER ENDS]CHRIS STETKIEWICZ: Youre listening to Ideas, a Microsoft Research Podcast that dives deep into the world of technology research and the profound questions behind the code. In this series, well explore the technologies that are shaping our future and the big ideas that propel them forward.[MUSIC FADES]Im your guest host, Chris Stetkiewicz. Today, Im talking to Akshay Nambi. Akshay is a principal researcher at Microsoft Research. His work lies at the intersection of systems, AI, and machine learning with a focus on designing, deploying, and scaling AI systems to solve compelling real-world problems. Akshays research extends across education, agriculture, transportation, and energy. He is currently working on enhancing the quality and reliability of AI systems by addressing critical challenges such as reasoning, grounding, and managing complex queries.Akshay, welcome to the podcast.AKSHAY NAMBI: Thanks for having me.STETKIEWICZ: Id like to begin by asking you to tell us your origin story. How did you get started on your path? Was there a big idea or experience that captured your imagination or motivated you to do what youre doing today?NAMBI: If I look back, my journey into research wasnt a straight line. It was more about discovering my passion through some unexpected opportunities and also finding purpose along the way. So before I started with my undergrad studies, I was very interested in electronics and systems. My passion for electronics, kind of, started when I was in school. I was more like an average student, not a nerd or not too curious, but I was always tinkering around, doing things, building stuff, and playing with gadgets and that, kind of, made me very keen on electronics and putting things together, and that was my passion. But sometimes things dont go as planned. So I didnt get into the college which I had hoped to join for electronics, so I ended up pursuing computer science, which wasnt too bad either. So during my final year of bachelors, I had to do a final semester project, which turned out to be a very pivotal moment. And thats when I got to know this institute called Indian Institute of Science (IISc), which is a top research institute in India and also globally. And I had a chance to work on a project there. And it was my first real exposure to open-ended research, right, so I remember where we were trying to build a solution that helped to efficiently construct an ontology for a specific domain, which simply means that we were building systems to help users uncover relationships in the data and allow them to query it more efficiently, right. And it was super exciting for me to design and build something new. And that experience made me realize that I wanted to pursue research further. And right after that project, I decided to explore research opportunities, which led me to join Indian Institute of Science again as a research assistant.STETKIEWICZ: So what made you want to take the skills you were developing and apply them to a research career?NAMBI: So interestingly when I joined IISc, the professor I worked with specialized in electronics, so things come back, so something I had always been passionate about. And I was the only computer science graduate in the lab at that time with others being electronic engineers, and I didnt even know how to solder. But the lab environment was super encouraging, collaborative, so I, kind of, caught up very quickly. In that lab, basically, I worked on several projects in the emerging fields of embedded device and energy harvesting systems. Specifically, we were designing systems that could harvest energy from sources like sun, hydro, and even RF (radio frequency) signals. And my role was kind of twofold. One, I designed circuits and systems to make energy harvesting more efficient so that you can store this energy. And then I also wrote programs, software, to ensure that the harvested energy can be used efficiently. For instance, as we harvest some of this energy, you want to have your programs run very quickly so that you are able to sense the data, send it to the server in an efficient way. And one of the most exciting projects I worked during that time was on data-driven agriculture. So this was back in 2008, 2009, right, where we developed an embedded system device with sensors to monitor the agricultural fields, collecting data like soil moisture, soil temperature. And that was sent to the agronomists who were able to analyze this data and provide feedback to farmers. In many remote areas, still access to power is a huge challenge. So we used many of the technologies we were developing in the lab, specifically energy harvesting techniques, to power these sensors and devices in the rural farms, and thats when I really got to see firsthand how technology could help peoples lives, particularly in rural settings. And thats what, kind of, stood out in my experience at IISc, right, was that it was [the] end-to-end nature of the work. And it was not just writing code or designing circuits. It was about identifying the real-world problems, solving them efficiently, and deploying solutions in the field. And this cemented my passion for creating technology that solves real-world problems, and thats what keeps me driving even today.STETKIEWICZ: And as youre thinking about those problems that you want to try and solve, where did you look for, for inspiration? It sounds like some of these are happening right there in your home.NAMBI: Thats right. Growing up and living in India, Ive been surrounded by these, kind of, many challenges. And these are not distant problems. These are right in front of us. And some of them are quite literally outside the door. So being here in India provides a unique opportunity to tackle some of the pressing real-world challenges in agriculture, education, or in road safety, where even small advancements can create significant impact.STETKIEWICZ: So how would you describe your research philosophy? Do you have some big goals that guide you?NAMBI: Right, as I mentioned, right, my research philosophy is mainly rooted in solving real-world problems through end-to-end innovation. For me, research is just not about pushing the boundaries of the knowledge. Its about ensuring that these advancements translate to meaningful impact on the ground, right. So, yes, the big goals that guide most of my work is twofold. One, how do we build technology thats scaled to benefit large populations? And two, at the same time, Im motivated by the challenge of tackling complex problems. That provides opportunity to explore, learn, and also create something new. And thats what keeps me excited.STETKIEWICZ: So lets talk a little bit about your journey at Microsoft Research. I know you began as an intern, and some of the initial work you did was focused on computer vision, road safety, energy efficiency. Tell us about some of those projects.NAMBI: As I was nearing the completion of my PhD, I was eager to look for opportunities in industrial labs, and Microsoft Research obviously stood out as an exciting opportunity. And additionally, the fact that Microsoft Research India was in my hometown, Bangalore, made it even more appealing. So when I joined as an intern, I worked together with Venkat Padmanabhan, who now leads the lab, and we started this project called HAMS, which stands for Harnessing Automobiles for Safety. As you know, road safety is a major public health issue globally, responsible for almost 1.35 million fatalities annually and with the situation being even more severe in countries like India. For instance, there are estimates that theres a life lost on the road every four minutes in India. When analyzing the factors which affect road safety, we saw mainly three elements. One, the vehicle. Second, the infrastructure. And then the driver. Among these, the driver plays the most critical role in many incidents, whether its over-speeding, driving without seat belts, drowsiness, fatigue, any of these, right. And this realization motivated us to focus on driver monitoring, which led to the development of HAMS. In a nutshell, HAMS is basically a smartphone-based system where youre mounting your smartphone on a windshield of a vehicle to monitor both the driver and the driving in real time with the goal of improving road safety. Basically, it observes key aspects such as where the driver is looking, whether they are distracted or fatigued[1], while also considering the external driving environment, because we truly believe to improve road safety, we need to understand not just the drivers action but also the context in which they are driving. For example, if the smartphones accelerometer detects sharp braking, the system would automatically check the distance to the vehicle in the front using the rear camera and whether the driver was distracted or fatigued using the front camera. And this holistic approach ensures a more accurate and comprehensive assessment of the driving behavior, enabling a more meaningful feedback.STETKIEWICZ: So that sounds like a system thats got several moving parts to it. And I imagine you had some technical challenges you had to deal with there. Can you talk about that?NAMBI: One of our guiding principles in HAMS was to use commodity, off-the-shelf smartphone devices, right. This should be affordable, in the range of $100 to $200, so that you can just take out regular smartphones and enable this driver and driving monitoring. And that led to handling several technical challenges. For instance, we had to develop efficient computer vision algorithms that could run locally on the device with cheap smartphone processing units while still performing very well at low-light conditions. We wrote multiple papers and developed many of the novel algorithms which we implemented on very low-cost smartphones. And once we had such a monitoring system, right, you can imagine theres several deployment opportunities, starting from fleet monitoring to even training new drivers, right. However, one application we hadnt originally envisioned but turned out to be its most impactful use case even today is automated drivers license testing. As you know, before you get a license, a driver is supposed to pass a test, but what happens in many places, including India, is that licenses are issued with very minimal or no actual testing, leading to unsafe and untrained drivers on the road. At the same time as we were working on HAMS, Indian government were looking at introducing technology to make testing more transparent and also automated. So we worked with the right set of partners, and we demonstrated to the government that HAMS could actually completely automate the entire license testing process. So we first deployed this system in Dehradun RTO (Regional Transport Office)which is the equivalent of a DMV in the USin 2019, working very closely with RTO officials to define what should be some of the evaluation criteria, right. Some of these would be very simple like, oh, is it the same candidate who is taking the test who actually registered for the test, right? And whether they are wearing seat belts. Did they scan their mirrors before taking a left turn and how well they performed in tasks like reverse parking and things like that.STETKIEWICZ: So whats been the government response to that? Have they embraced it or deployed it in a wider extent?NAMBI: Yes, yes. So after the deployment in Dehradun in 2019, we actually open sourced the entire HAMS technology and our partners are now working with several state governments and scaled HAMS to several states in India. And as of today, we have around 28 RTOs where HAMS is actually being deployed, and the pass rate of such license test is just 60% as compared to 90-plus percent with manual testing. Thats the extensive rigor the system brings in. And now what excites me is after nearly five years later, we are now taking the next step in this project where we are now evaluating the long-term impact of this intervention on driving behavior and road safety. So we are collaborating with Professor Michael Kremer, who is a Nobel laureate and professor at University of Chicago, and his team to study how this technology has influenced driving patterns and accident rates over time. So this focus on closing the loop and moving beyond just deployment in the field to actually measuring the real impact, right, is something that truly excites me and that makes research at Microsoft is very unique. And that is actually one of the reasons why I joined Microsoft Research as a full-time after my internship, and this unique flexibility to work on real-world problems, develop novel research ideas, and actually collaborate with partners both internally and externally to deploy at scale is something that is very unique here.STETKIEWICZ: So have you actually received any evidence that the project is working? Is driving getting safer?NAMBI: Yes, these are very early analysis, and there are very positive insights we are getting from that. Soon we will be releasing a white paper on our study on this long-term impact.STETKIEWICZ: Thats great. I look forward to that one. So youve also done some interesting work involving the Internet of Things, with an emphasis on making it more reliable and practical. So for those in our audience who may not know, the Internet of Things, or IoT, is a network that includes billions of devices and sensors in things like smart thermostats and fitness trackers. So talk a little bit about your work in this area.NAMBI: Right, so IoT, as you know, is already transforming several industries with billions of sensors being deployed in areas like industrial monitoring, manufacturing, agriculture, smart buildings, and also air pollution monitoring. And if you think about it, these sensors provide critical data that businesses rely for decision making. However, a fundamental challenge is ensuring that the data collected from these sensors is actually reliable. If the data is faulty, it can lead to poor decisions and inefficiencies. And the challenge is that these sensor failures are always not obvious. What I mean by that is when a sensor stops working, it always doesnt stop sending data, but it often continues to send some data which appear to be normal. And thats one of the biggest problems, right. So detecting these errors is non-trivial because the faulty sensors can mimic real-world working data, and traditional solutions like deploying redundant sensors or even manually inspecting them are very expensive, labor intensive, and also sometimes infeasible, especially for remote deployments. Our goal in this work was to develop a simple and efficient way to remotely monitor the health of the IoT sensors. So what we did was we hypothesized that most sensor failures occurred due to the electronic malfunctions. It could be either due to short circuits or component degradation or due to environmental factors such as heat, humidity, or pollution. Since these failures originate within the sensor hardware itself, we saw an opportunity to leverage some of the basic electronic principles to create a novel solution. The core idea was to develop a way to automatically generate a fingerprint for each sensor. And by fingerprint, I mean the unique electrical characteristic exhibited by a properly working sensor. We built a system that could devise these fingerprints for different types of sensors, allowing us to detect failures purely based on the sensors internal characteristics, that is the fingerprint, and even without looking at the data it produces. Essentially what it means now is that we were able to tag each sensor data with a reliability score, ensuring verifiability.STETKIEWICZ: So how does that technology get deployed in the real world? Is there an application where its being put to work today?NAMBI: Yes, this technology, we worked together with Azure IoT and open-sourced it where there were several opportunities and several companies took the solution into their systems, including air pollution monitoring, smart buildings, industrial monitoring. The one which I would like to talk about today is about air pollution monitoring. As you know, air pollution is a major challenge in many parts of the world, especially in India. And traditionally, air quality monitoring relies on these expensive fixed sensors, which provide limited coverage. On the other hand, there is a rich body of work on low-cost sensors, which can offer wider deployment. Like, you can put these sensors on a bus or a vehicle and have it move around the entire city, where you can get much more fine-grained, accurate picture on the ground. But these are often unreliable because these are low-cost sensors and have reliability issues. So we collaborated with several startups who were developing these low-cost air pollution sensors who were finding it very challenging to gain trust because one of the main concerns was theaccuracy of the data from low-cost sensors. So our solution seamlessly integrated with these sensors, which enabled verification of the data quality coming out from these low-cost air pollution sensors. So this bridged the trust gap, allowing government agenciesto initiate large-scale pilots using low-cost sensors for fine-grain air-quality monitoring.STETKIEWICZ: So as were talking about evolving technology, large language models, or LLMs, are also enabling big changes, and theyre not theoretical. Theyre happening today. And youve been working on LLMs and their applicability to real-world problems. Can you talk about your work there and some of the latest releases?NAMBI: So when ChatGPT was first released, I, like many people, was very skeptical. However, I was also curious both of how it worked and, more importantly, whether it could accelerate solutions to real-world problems. That led to the exploration of LLMs in education, where we fundamentally asked this question, can AI help improve educational outcomes? And this was one of the key questions which led to the development of Shiksha copilot, which is a genAI-powered assistant designed to support teachers in their daily work, starting from helping them to create personalized learning experience, design assignments, generate hands-on activities, and even more. Teachers today universally face several challenges, from time management to lesson planning. And our goal with Shiksha was to empower them to significantly reduce the time spent on this task. For instance, lesson planning, which traditionally took about 60 minutes, can now be completed in just five minutes using the Shiksha copilot. And what makes Shiksha unique is that its completely grounded in the local curriculum and the learning objectives, ensuring that the AI-generated content aligns very well with the pedagogical best practices. The system actually supports multilingual interactions, multimodal capabilities, and also integration with external knowledge base, making it very highly adaptable for different curriculums. Initially, many teachers were skeptical. Some feared this would limit their creativity. However, as they began starting to use Shiksha, they realized that it didnt replace their expertise, but rather amplified it, enabling them to do work faster and more efficiently.STETKIEWICZ: So, Akshay, the last time you and I talked about Shiksha copilot, it was very much in the pilot phase and the teachers were just getting their hands on it. So it sounds like, though, youve gotten some pretty good feedback from them since then.NAMBI: Yes, so when we were discussing, we were doing this six-month pilot with 50-plus teachers where we gathered overwhelming positive feedback on how technologies are helping teachers to reduce time in their lesson planning. And in fact, they were using the system so much that they really enjoyed working with Shiksha copilot where they were able to do more things with much less time, right. And with a lot of feedback from teachers, we have improved Shiksha copilot over the past few months. And starting this academic year, we have already deployed Shiksha to 1,000-plus teachers in Karnataka. This is with close collaboration with our partners in with the Sikshana Foundation and also with the government of Karnataka. And the response has been already incredibly encouraging. And looking ahead, we are actually focusing on again, closing this loop, right, and measuring the impact on the ground, where we are doing a lot of studies with the teachers to understand not just improving efficiency of the teachers but also measuring how AI-generated content enriched by teachers is actually enhancing student learning objectives. So thats the study we are conducting, which hopefully will close this loop and understand our original question that, can AI actually help improve educational outcomes?STETKIEWICZ: And is the deployment primarily in rural areas, or does it include urban centers, or whats the target?NAMBI: So the current deployment with 1,000 teachers is a combination of both rural and urban public schools. These are covering both English medium and Kannada medium teaching schools with grades from Class 5 to Class 10.STETKIEWICZ: Great. So Shiksha was focused on helping teachers and making their jobs easier, but I understand youre also working on some opportunities to use AI to help students succeed. Can you talk about that?NAMBI: So as you know, LLMs are still evolving and inherently they are fragile, and deploying them in real-world settings, especially in education, presents a lot of challenges. With Shiksha, if you think about it, teachers remain in control throughout the interaction, making the final decision on whether to use the AI-generated content in the classroom or not. However, when it comes to AI tutors for students, the stakes are slightly higher, where we need to ensure the AI doesnt produce incorrect answers, misrepresent concepts, or even mislead explanations. Currently, we are developing solutions to enhance accuracy and also the reasoning capabilities of these foundational models, particularly solving math problems. This represents a major step towards building AI systems thats much more holistic personal tutors, which help student understanding and create more engaging, effective learning experience.STETKIEWICZ: So youve talked about working in computer vision and IoT and LLMs. What do those areas have in common? Is there some thread that weaves through the work that youre doing?NAMBI: Thats a great question. As a systems researcher, Im quite interested in this end-to-end systems development, which means that my focus is not just about improving a particular algorithm but also thinking about the end-to-end system, which means that I, kind of, think about computer vision, IoT, and even LLMs as tools, where we would want to improve them for a particular application. It could be agriculture, education, or road safety. And then how do you think this holistically to come up with the best efficient system that can be deployed at population scale, right. I think thats the connecting story here, that how do you have this systemic thinking which kind of takes the existing tools, improves them, makes it more efficient, and takes it out from the lab to real world.STETKIEWICZ: So youre working on some very powerful technology that is creating tangible benefits for society, which is your goal. At the same time, were still in the very early stages of the development of AI and machine learning. Have you ever thought about unintended consequences? Are there some things that could go wrong, even if we get the technology right? And does that kind of thinking ever influence the development process?NAMBI: Absolutely. Unintended consequences are something I think about deeply. Even the most well-designed technology can have these ripple effects that we may not fully anticipate, especially when we are deploying it at population scale. For me, being proactive is one of the key important aspects. This means not only designing the technology at the lab but actually also carefully deploying them in real world, measuring its impact, and working with the stakeholders to minimize the harm. In most of my work, I try to work very closely with the partner team on the ground to monitor, analyze, how the technology is being used and what are some of the risks and how can we eliminate that. At the same time, I also remain very optimistic. Its also about responsibility. If we are able to embed societal values, ethics, into the design of the system and involve diverse perspectives, especially from people on the ground, we can remain vigilant as the technology evolves and we can create systems that can truly deliver immense societal benefits while addressing many of the potential risks.STETKIEWICZ: So weve heard a lot of great examples today about building technology to solve real-world problems and your motivation to keep doing that. So as you look ahead, where do you see your research going next? How will people be better off because of the technology you develop and the advances that they support?NAMBI: Yeah, Im deeply interested in advancing AI systems that can truly assist anyone in their daily tasks, whether its providing personalized guidance to a farmer in a rural village, helping a student get instant 24 by 7 support for their learning doubts, or even empowering professionals to work more efficiently. And to achieve this, my research is focusing on tackling some of the fundamental challenges in AI with respect to reasoning and reliability and also making sure that AI is more context aware and responsive to evolving user needs. And looking ahead, I envision AI as not just an assistant but also as an intelligent and equitable copilot seamlessly integrated into our everyday life, empowering individuals across various domains.STETKIEWICZ: Great. Well, Akshay, thank you for joining us on Ideas. Its been a pleasure.[MUSIC]NAMBI: Yeah, I really enjoyed talking to you, Chris. Thank you.STETKIEWICZ: Till next time.[MUSIC FADES]

Large language models (LLMs) are increasingly being deployed on edge deviceshardware that processes data locally near the data source, such as smartphones, laptops, and robots. Running LLMs on these devices supports advanced AI and real-time services, but their massive size, with hundreds of millions of parameters, requires significant memory and computational power, limiting widespread adoption. Low-bit quantization, a technique that compresses models and reduces memory demands, offers a solution by enabling more efficient operation.Recent advances in low-bit quantization have made mixed-precision matrix multiplication (mpGEMM) viable for LLMs. This deep learning technique allows data of the same or different formats to be multiplied, such as int8*int1, int8*int2, or FP16*int4. By combining a variety of precision levels, mpGEMM strikes a balance among speed, memory efficiency, and computational accuracy.However, most hardware supports only symmetric computationsoperations on data of similar formatscreating challenges for mixed-precision calculations during General Matrix Multiplication (GEMM), a critical operation for LLMs. Overcoming these hardware limitations is essential to fully benefit from mpGEMM and support asymmetrical computations.To unlock the potential of low-bit quantization on resource-constrained edge devices, hardware must natively support mpGEMM. To address this, we developed the following three approaches for computing kernels and hardware architectures:Ladder data type compiler: Supports various low-precision data types by converting unsupported types into hardware-compatible ones without data loss, while also generating high-performance conversion code.T-MAC mpGEMM library: Implements GEMM using a lookup table (LUT) approach, eliminating multiplications to significantly reduce computational overhead. Optimized for diverse CPUs, T-MAC delivers several times the speed of other libraries.LUT Tensor Core hardware architecture: Introduces a cutting-edge design for next-generation AI hardware, tailored for low-bit quantization and mixed-precision computations.The following sections describe these techniques in detail.Ladder: Bridging the gap between custom data and hardware limitsCutting-edge hardware accelerators, such as GPUs, TPUs, and specialized chips, are designed to speed up computationally intensive tasks like deep learning by efficiently handling large-scale operations. These accelerators now integrate lower-bit computing units, such as FP32, FP16, and even FP8, into their architectures.However, constraints in chip area and hardware costs limit the availability of these units for standard data types. For instance, the NVIDIA V100 Tensor Core GPU supports only FP16, while the A100 supports int2, int4, and int8 but not newer formats like FP8 or OCP-MXFP. Additionally, the rapid development of LLMs often outpaces hardware upgrades, leaving many new data formats unsupported and complicating deployment.Additionally, while hardware accelerators may lack direct support for custom data types, their memory systems can convert these types into fixed-width data blocks that store any data format. For instance, NF4 tensors can be converted into FP16 or FP32 for floating-point operations.Building on these insights, we developed the Ladder data type compiler, a method to separate data storage from computation, enabling broader support for custom data types. It bridges the gap between emerging custom data formats with the precision types supported by current hardware.Ladder offers a flexible system for converting between algorithm-specific and hardware-supported data types without data loss. For low-bit applications, it optimizes performance by translating low-bit data into the most efficient formats for the hardware being used. As shown in Figure 1, this includes mapping low-bit computations to supported instructions and efficiently managing data storage across the memory hierarchy.Figure 1: The Ladder architectureEvaluating LadderEvaluations of Ladder on NVIDIA and AMD GPUs show that it outperforms existing deep neural network (DNN) compilers for natively supported data types. It also handles custom data types not supported by GPUs, achieving speedups of up to 14.6 times.As the first system to support custom low-precision data types for running DNNs on modern hardware accelerators, Ladder provides researchers with flexibility in optimizing data types. It also enables hardware developers to support a wider range of data types without requiring hardware modifications.T-MAC: Table-lookup for mpGEMM without multiplicationDeploying low-bit quantized LLMs on edge devices often requires dequantizing models to ensure hardware compatibility. However, this approach has two major drawbacks:Performance: Dequantization overhead can result in poor performance, negating the benefits of low-bit quantization.Development: Developers must redesign data layouts and kernels for different mixed precisions.To address these challenges, we introduce T-MAC, a novel LUT-based method that enables mpGEMM without dequantization or multiplication.T-MAC replaces traditional multiplication operations with bit-wise table lookups, offering a unified and scalable solution for mpGEMM. It incorporates techniques to reduce the size of tables and store them directly on the chip, minimizing the overhead of accessing data from memory. By eliminating dequantization and lowering computational costs, T-MAC enables efficient inference of low-bit LLMs on resource-constrained edge devices. Figure 2 illustrates T-MACs architecture.Figure 2. Overview of the T-MAC systemEvaluating T-MACPerformance evaluations of T-MAC on low-bit models demonstrated substantial benefits in efficiency and speed. On the Surface Laptop 7 with the Qualcomm Snapdragon X Elite chipset, T-MAC achieved:48 tokens per second for the 3B BitNet-b1.58 model30 tokens per second for the 2-bit 7B Llama model20 tokens per second for the 4-bit 7B Llama modelThese speeds far exceed average human reading rates, outperforming llama.cpp by 45 times and doubling the speed of a dedicated NPU accelerator.Even on lower-end devices like the Raspberry Pi 5, T-MAC made it possible for the 3B BitNet-b1.58 model to generate 11 tokens per second. It also proved highly power-efficient, matching llama.cpps generation rate while using only 1/4 to 1/6 of the CPU cores.These results establish T-MAC as a practical solution for deploying LLMs on edge devices with standard CPUs, without relying on GPUs or NPUs. T-MAC allows LLMs to run efficiently on resource-constrained devices, expanding their applicability across a wider range of scenarios.LUT Tensor Core: Driving hardware for mpGEMMWhile T-MAC and Ladder optimize mpGEMM on existing CPU and GPU architectures, improving computational efficiency, they cannot match the performance of dedicated hardware accelerators with built-in LUT support. Achieving significant improvements in performance, power, and area (PPA) requires overcoming four key challenges:Table precompute and storage: Precomputing and storing LUTs add overhead, increasing area usage, latency, and storage requirements, which can reduce overall efficiency gains.Bit-width flexibility: Hardware must support various precision levels, such as int4/2/1 for weights and FP16/8 or int8 for activations, along with their combinations. This flexibility is crucial for accommodating diverse model architectures and use cases.LUT tiling shape: Inefficient tiling shapes can raise storage costs and limit reuse opportunities, adversely affecting performance and efficiency.Instruction and compilation: LUT-based mpGEMM requires a new instruction set. Existing compilation stacks, designed for standard GEMM hardware, may not optimally map and schedule these instructions, complicating integration with LLM inference software.In response, we developed LUT Tensor Core, a software-hardware codesign for low-bit LLM inference. To address precomputation overhead in conventional LUT-based methods, we introduce techniques like software-based DFG transformation, operator fusion, and table symmetrization to optimize table precomputation and storage. Additionally, we propose a hardware design with an elongated tiling shape to support table reuse and a bit-serial design to handle various precision combinations in mpGEMM.To integrate with existing GPU microarchitectures and software stacks, we extended the MMA instruction set, added new LMMA instructions, and developed a cuBLAS-like software stack for easy integration into existing DNN frameworks. We also created a compiler for end-to-end execution planning on GPUs with LUT Tensor Core. This design and workflow, illustrated in Figure 3, enabled the quick and seamless adoption of LUT Tensor Core.Figure 3. The LUT Tensor Core workflowEvaluating LUT Tensor CoreTesting LUT Tensor Core on low-bit LLMs, such as BitNet and Llama, showed significant performance gains, achieving 6.93 times the inference speed while using just 38.3% of the area of a traditional Tensor Core. With nearly identical model accuracy, this results in a 20.9-fold increase in computational density and an 11.2-fold boost in energy efficiency. As AI models grow in scale and complexity, LUT Tensor Core enables low-bit LLMs to be applied in new and diverse scenarios.We believe the LUT technique could drive a paradigm shift in AI model inference. Traditional methods rely on multiplication and accumulation operations, whereas LUT implementations provide higher transistor density, greater throughput per chip area, lower energy costs, and better scalability. As large models adopt low-bit quantization, the LUT method could become the standard for system and hardware design, advancing the next generation of AI hardware innovation.Unlocking new possibilities for embodied AILow-bit quantization improves the efficiency of running large models on edge devices while also enabling model scaling by reducing the bits used to represent each parameter. This scaling enhances model capabilities, generality, and expressiveness, as shown by the BitNet model, which starts with a low-bit configuration and expands.Technologies like T-MAC, Ladder, and LUT Tensor Core provide solutions for running low-bit quantized LLMs, supporting efficient operation across edge devices and encouraging researchers to design and optimize LLMs using low-bit quantization. By reducing memory and computational demands, low-bit LLMs could power embodied AI systems, such as robots, enabling dynamic perception and real-time environmental interaction.T-MAC (opens in new tab) and Ladder (opens in new tab) are open source and available on GitHub. We invite you to test and explore these innovations in AI technology with Microsoft Research.Spotlight: blog postGraphRAG auto-tuning provides rapid adaptation to new domainsGraphRAG uses LLM-generated knowledge graphs to substantially improve complex Q&A over retrieval-augmented generation (RAG). Discover automatic tuning of GraphRAG for new datasets, making it more accurate and relevant.Read moreOpens in a new tab Opens in a new tab

In this edition:We introduce FLAVARS, a multimodal foundation language and vision alignment model for remote sensing; Managed-retention memory, a new class of memory which is more optimized to store key data structures for AI inference workloads; and Enhanced detection of macular telangiectasia type 2 (MacTel 2) using self-supervised learning and ensemble models.We present a new approach to generalizing symbolic automata, which brings together a variety of classic automata and logics in a unified framework with all the necessary ingredients to support symbolic model checking moduloA.And we invite you to join an upcoming workshop: LLM4Eval@WSDM 2025: Large Language Models for Evaluation in Information Retrieval. LLM4Eval is a promising technique in the areas of automated judgments, natural language generation, and retrieval augmented generation (RAG) systems. Researchers from Microsoft and experts from industry and academia will explore this technique at an interactive workshop on Friday, March 14, in Hanover, Germany.NEW RESEARCHIn the field of remote sensing, imagery is generally dense with objects and visual content which can vary regionally across the globe. This creates a need for vision-language datasets to be highly detailed when describing imagery, and for pretraining to better balance visual task performance while retaining the ability to perform zero-shot classification and image-text retrieval.One strategy is to combine paired satellite images and text captions for pretraining performant encoders for downstream tasks. However, while contrastive image-text methods like CLIP enable vision-language alignment and zero-shot classification ability, CLIPs vision-only downstream performance tends to degrade compared to image-only pretraining, such as Masked Autoencoders (MAE).To better approach multimodal pretraining for remote sensing, researchers from Microsoft propose a pretraining method that combines the best of both contrastive learning and masked modeling, along with geospatial alignment via contrastive location encoding, in the recent paper: FLAVARS: A Multimodal Foundational Language and Vision Alignment Model for Remote Sensing. The research shows that FLAVARS significantly outperforms a baseline of SkyCLIP for vision-only tasks such as KNN classification and semantic segmentation, +6% mIOU on SpaceNet1, while retaining the ability to perform zero-shot classification, unlike MAE pretrained methods.Read the paperNEW RESEARCHAI clusters today are one of the major uses of high bandwidth memory (HBM), a high-performance type of computer memory. However, HBM is suboptimal for AI inference workloads for several reasons. Analysis shows that HBM is overprovisioned on write performance, underprovisioned on density and read bandwidth, and has significant energy-per-bit overhead. It is also expensive, with lower yield than DRAM due to manufacturing complexity.In a recent paper: Managed-Retention Memory: A New Class of Memory for the AI Era, researchers from Microsoft propose a memory class which is more optimized to store key data structures for AI inference workloads. The paper makes the case that MRM may finally provide a path to viability for technologies that were originally proposed to support storage class memory (SCM). These technologies traditionally offered long-term persistence (10+ years) but provided poor IO performance and/or endurance. MRM makes different trade-offs, and by understanding the workload IO patterns, MRM foregoes long-term data retention and write performance for better potential performance on the metrics important for AI inference.Read the paperNEW RESEARCHMacular telangiectasia type 2 (MacTel) is a retinal disease that is challenging to diagnose. While increased awareness has led to improved diagnostic outcomes, MacTel diagnosis relies significantly upon a multimodal image set and the expertise of clinicians familiar with the disease. Optical coherence tomography (OCT) imaging has emerged as a valuable tool for the diagnosis and monitoring of various retinal diseases.With the increasing integration of OCT into clinical practice, deep learning models may be able to achieve accurate MacTel prediction comparable to that of retinal specialists, even when working with limited data.Researchers from Microsoft and external colleagues address this challenge in a recent paper: Enhanced Macular Telangiectasia Type 2 Detection: Leveraging Self-Supervised Learning and Ensemble Models. Published in the journal of Ophthalmology Science, the paper focuses on the accurate classification of macular telangiectasia type 2 using OCT images, with the overarching goal of facilitating early and precise detection of this neurodegenerative disease.The researchers present results leveraging self-supervised learning and ensemble models, showing their approach improves both MacTel classification accuracy and interpretability when compared to the use of individual models. Ensemble models exhibited superior agreement with the assessments of the most experienced individual human experts, as well as the ensemble of human experts.Read the paperMicrosoft research podcastCollaborators: Silica in space with Richard Black and Dexter GreeneCollege freshman Dexter Greene and Microsoft research manager Richard Black discuss how technology that stores data in glass is supporting students as they expand earlier efforts to communicate what it means to be human to extraterrestrials.Listen nowOpens in a new tab NEW RESEARCHSymbolic automata are finite state automata that support potentially infinite alphabets, such as the set of rational numbers, generally applied to regular expressions and languages over finite words. In symbolic automata (or automata moduloA), an alphabet is represented by an effective Boolean algebraA, supported by a decision procedure for satisfiability. Regular languages over infinite words (so called -regular languages) have a rich history paralleling that of regular languages over finite words, with well-known applications to model checking via Bchi automata and temporal logics.In a recent paper: Symbolic Automata: Omega-Regularity Modulo Theories, researchers from Microsoft generalize symbolic automata to support -regular languages viatransition termsandsymbolic derivatives. This brings together a variety of classic automata and logics in a unified framework that provides all the necessary ingredients to support symbolic model checking moduloA.Read the paperEVENTLLMs have shown increasing task-solving abilities not present in smaller models. Using LLMs for automated evaluation (LLM4Eval) is a promising technique in the areas of automated judgments, natural language generation, and retrieval augmented generation (RAG) systems.Join researchers from Microsoft and experts from industry and academia for a discussion on using LLMs for evaluation in information retrieval at LLM4Eval Workshop WSDM 2025 (opens in new tab), March 14, 2025, in Hanover, Germany.This interactive workshop will cover automated judgments, RAG pipeline evaluation, altering human evaluation, robustness, and trustworthiness of LLMs for evaluation in addition to their impact on real-world applications. The organizers believe that the information retrieval community can significantly contribute to this growing research area by designing, implementing, analyzing, and evaluating various aspects of LLMs with applications to LLM4Eval tasks.Learn more about the workshopMicrosoft Research | In case you missed itMicrosoft Team Uses Diffusion Model For Materials ScienceJanuary 21, 2025Finding a new material for a target application is like finding a needle in a haystack, write the authors of a blog post at Microsoft, where they have been working on just such a program, something called, aptly, MatterGen. Microsoft AutoGen v0.4: A turning point toward more intelligent AI agents for enterprise developersJanuary 18, 2025The world of AI agents is undergoing a revolution, and Microsofts release of AutoGen v0.4 this week marked a significant leap forward in this journey. Positioned as a robust, scalable and extensible framework, AutoGen represents Microsofts latest attempt to address the challenges of building multi-agent systems for enterprise applications. 2 AI breakthroughs unlock new potential for health and scienceJanuary 17, 2025Two new research papers published this week in scientific journals, one in Nature and one in Nature Machine Intelligence, show how generative AI foundation models can exponentially speed up scientific discovery of new materials and help doctors access and analyze radiology results faster. ChatGPT gets proactive with 'Tasks'January 15, 2025Good morning, AI enthusiasts. OpenAIs AI agent era just got its unofficial start with ChatGPT gaining the ability to schedule and manage daily tasks. With Tasks rolling out and mysterious Operator whispers in the air, is OpenAI finally ready to move from chatbots to full-on autonomous assistants? Mayo Clinic and Microsoft partner to advance generative AI in radiologyJanuary 15, 2025The Mayo Clinic is seeking to advance the use of generative artificial intelligence in imaging through a new collaboration with Microsoft Research. The duo made the announcement during the 43rd Annual J.P. Morgan Healthcare Conference taking place now in San Francisco. View more news and awards Opens in a new tab

Transcript[TEASER][MUSIC PLAYS UNDER DIALOGUE]SHAN LU: I remember, you know, those older days myself, right. That is really, like, I have this struggle that I feel like I can do better. I feel like I have ideas to contribute. But just for whatever reason, right, it took me forever to learn something which I feel like its a very mechanical thing, but it just takes me forever to learn, right. And then now actually, I see this hope, right, with AI. You know, a lot of mechanical things that can actually now be done in a much more automated way, you know, by AI, right. So then now truly, you know, my daughter, many girls, many kids out there, right, whatever, you know, they are good at, their creativity, itll be much easier, right, for them to contribute their creativity to whatever discipline they are passionate about.[TEASER ENDS]GRETCHEN HUIZINGA:Ideas, a Microsoft Research Podcast that dives deep into the world of technology research and the profound questions behind the code. Im Gretchen Huizinga. In this series, well explore the technologies that are shaping our future and the big ideas that propel them forward.[MUSIC FADES]Today Im talking to Shan Lu, a senior principal research manager at Microsoft Research and a computer science professor at the University of Chicago. Part of the Systems Research Group, Shan and her colleagues are working to make our computer systems, and I quote, secure, scalable, fault tolerant, manageable, fast, and efficient. Thats no small order, so Im excited to explore the big ideas behind Shans influential research and find out more about her reputation as a bug bounty hunter. Shan Lu, welcome to Ideas!SHAN LU: Thank you.HUIZINGA: So I like to start these episodes with what Ive been calling the research origin story, and you have a unique, almost counterintuitive, story about what got you started in the field of systems research. Would you share that story with our listeners?LU: Sure, sure. Yeah. I grew up fascinating that I will become mathematician. I think I was good at math, and at some point, actually, until, I think, I entered college, I was still, you know, thinking about, should I do math? Should I do computer science? For whatever reason, I think someone told me, you know, doing computer science will help you; its easier to get a job. And I reluctantly pick up computer science major. And then there was a few years in my college, I had a really difficult time for programming. And I also remember that there was, like, I spent a lot of time learning one languagewe started with Pascaland I feel like I finally know what to do and then theres yet another language, C, and another class, Java. And I remember, like, the teacher will ask us to do a programming project, and there are times I dont even, I just dont know how to get started. And I remember, at that time, in my class, I think there were we only had like four girls taking this class that requires programming in Java, and none of us have learned Java before. And when we ask our classmates, when we ask the boys, they just naturally know what to do. It was really, really humiliating. Embarrassing. I had the feeling that, I felt like Im just not born to be a programmer. And then, I came to graduate school. I was thinking about, you know, what kind of research direction I should do. And I was thinking that, oh, maybe I should do theory research, like, you know, complexity theory or something. You know, after a lot of back and forth, I met my eventual adviser. She was a great, great mentor to me, and she told me that, hey, Shan, you know, my group is doing research about finding bugs in software. And she said her group is doing system research, and she said a lot of current team members are all great programmers, and as a result, they are not really well-motivated [LAUGHS] by finding bugs in software!HUIZINGA: Interesting.LU: And then she said, you are really motivated, right, by, you know, getting help to developers, to help developers finding bugs in their software, so maybe thats the research project for you. So thats how I got started.HUIZINGA: Well, lets go a little bit further on this mentor and mentors in general. As Dr. Seuss might say, every what has a who. So by that I mean an inspirational person or people behind every successful researchers career. And most often, theyre kind of big names and meaningful relationships, but you have another unique story on who has influenced you in your career, so why dont you tell us about the spectrum of people whove been influential in your life and your career?LU: Mm-hmm. Yeah, I mean, I think I mentioned my adviser, and shes just so supportive. And I remember, when I started doing research, I just felt like I seemed to be so far behind everyone else. You know, I felt like, how come everybody else knows how to ask, you know, insightful questions? And they, like, they know how to program really fast, bug free. And my adviser really encouraged me, saying, you know, there are background knowledge that you can pick up; you just need to be patient. But then there are also, like, you know how to do research, you know how to think about things, problem solving. And she encouraged me saying, Shan, youre good at that!HUIZINGA: Interesting!LU: Well, I dont know how she found out, and anyway, so she was super, super helpful.HUIZINGA: OK, so go a little further on this because I know you have others that have influence you, as well.LU: Yes. Yes, yes. And I think those, to be honest, Im a very emotional, sensitive person. I would just, you know, move the timeline to be, kind of, more recent. So I joined Microsoft Research as a manager, and theres something called Connect that, you know, people write down twice every year talking about what it is theyve been doing. So I was just checking, you know, my members in my team to see what they have been doing over the years just to just get myself familiar with them. And I remember I read several of them. I felt like I almost have tears in my eyes! Like, I realized, wow, like And just to give example, for Chris, Chris Hawblitzel, I read his Connect, and I saw that hes working on something called program verification. Its a very, very difficult problem, and [as an] outsider, you know, Ive read many of his papers, but when I read, you know, his own writing, and I realized, wow, you know, its almost two decades, right. Like, he just keeps doing these very difficult things. And I read his words about, you know, how his old approach has problems, how hes thinking about how to address that problem. Oh, I have an idea, right. And then spend multiple years to implement that idea and get improvement; find a new problem and then just find new solutions. And I really feel like, wow, Im really, really, like, I feel like this is, kind of, like a, you know, theres, how to say, a hero-ish story behind this, you know, this kind of goal, and youre willing to spend many years to keep tackling this challenging problem. And I just feel like, wow, Im so honored, you know, to be in the same group with a group of fighters, you know, determined to tackle difficult research problems.HUIZINGA: Yeah. And I think when you talk about it, its like this is a person that was working for you, a direct report. [LAUGHTER] And often, we think about our heroes as being the ones who mentored us, who taught us, who managed us, but yours is kind of 360! Its like LU: True!HUIZINGA: your heroes [are] above, beside and below.LU: Right. And I would just say that I have many other, you know, direct reports in my group, and I have, you know, for example, say a couple other my colleagues, my direct reports, Dan Ports and Jacob Nelson. And again, this is something like their story really inspired me. Like, they were, again, spent five or six years on something, and it looks like, oh, its close to the success of tech transfer, and then something out of their control happened. It happened because Intel decided to stop manufacturing a chip that their research relied on. And its, kind of, like the end of the world to them, HUIZINGA: Yeah.LU: and then they did not give up. And then, you know, like, one year later, they found a solution, you know, together with their product team collaborators.HUIZINGA: Wow.LU: And I still feel like, wow, you know, I feel so I feel like Im inspired every day! Like, Im so happy to be working together with, you know, all these great people, great researchers in my team.HUIZINGA: Yeah. Wow. So much of your work centers on this idea of concurrent systems and I want you to talk about some specific examples of this work next, but I think it warrants a little explication upfront for those people in the audience who dont spend all their time working on concurrent systems themselves. So give us a short 101 on concurrent systems and explain why the work you do matters to both the people who make it and the people who use it.LU: Sure. Yeah. So I think a lot of people may not realize so actually, the software were using every day, almost every software we use these days are concurrent. So the meaning of concurrent is that you have multiple threads of execution going on at the same time, in parallel. And then, when we go to a web browser, right, so its not just one rendering that is going on. There are actually multiple concurrent renderings that is going on. So the problem of writing for software developers to develop this type of concurrent system, a challenge is the timing. So because you have multiple concurrent things going on, its very difficult to manage and reason about, you know, what may happen first, what may happen second. And also, its, like, theres an inherent non-determinism in it. What happened first this time may happen second next time. So as a result, a lot of bugs are introduced by this. And it was a very challenging problem because I would say about 20 years ago, there was a shift. Like, in the older days, actually most of our software is written in a sequential way instead of a concurrent way. So, you know, a lot of developers also have a difficult time to shift their mindset from the sequential way of reasoning to this concurrent way of reasoning.HUIZINGA: Right. Well, and I think, from a users perspective, all you experience is what I like to call the spinning beachball of doom. Its like Ive asked something, and it doesnt want to give, so [LAUGHS] And this is, like, behind the scenes from a reasoning perspective of, how do we keep that from happening to our users? How do we identify the bugs? Which well get to in a second. Umm. Thanks for that. Your research now revolves around what I would call the big idea of learning from mistakes. And in fact, it all seems to have started with a paper that you published way back in 2008 called Learning from Mistakes: A Comprehensive Study on Real World Concurrency Bug Characteristics, and you say this strongly influenced your research style and approach. And by the way, Ill note that this paper received the Most Influential Paper Award in 2022 from ASPLOS, which is the Architectural Support for Programming Languages and Operating Systems. Huge mouthful. And it also has more than a thousand citations, so I dare say its influenced other researchers approach to research, as well. Talk about the big idea behind this paper and exactly how it informed your research style and approach today.LU: Mm-hmm. Yeah. So I think this, like, again, went back to the days that I, you know, my PhD days, I started working with my adviser, you know, YY (Yuanyuan Zhou). So at that time, there had been a lot of people working on bug finding, but then now when I think about it, people just magically say, hey, I want to look at this type of bug. Just magically, oh, I want to look at that type of bug. And then, my adviser at that time suggested to me, saying, hey, maybe, you know, actually take a look, right. At that time, as I mentioned, software was kind of shifting from sequential software to concurrent software, and my adviser was saying, hey, just take a look at those real systems bug databases, and see what type of concurrency bugs are actually there. You know, instead of just randomly saying, oh, I want to work on this type of bug.HUIZINGA: Oh, yeah.LU: And then also, of course, its not just look at it. Its not just like you read a novel or something, right. [LAUGHTER] And again, my adviser said, hey, Shan, right, you have this, you have a connection, natural connection, you know, with bugs and the developers who commit HUIZINGA: Who make them LU: Who make them! [LAUGHTER] So she said, you know, try to think about the patterns behind them, right. Try to think about whether you can generalize some HUIZINGA: Interesting LU: characteristics, and use that to guide peoples research in this domain. And at that time, we were actually thinking we dont know whether, you know, we can actually write a paper about it because traditionally you publish a paper, just say, oh, I have a new tool, right, which can do this and that. At that time in system conferences, people rarely have, you know, just say, heres a study, right. But we studied that, and indeed, you know, I had this thought that, hey, why I make a lot of mistakes. And when I study a lot of bugs, the more and more, I feel, you know, theres a reason behind it, right. Its like Im not the only dumb person in the world, right? [LAUGHTER] Theres a reason that, you know, theres some part of this language is difficult to use, right, and theres a certain type of concurrent reasoning, its just not natural to many people, right. So because of that, there are patterns behind these bugs. And so at that time, we were surprised that the paper was actually accepted. Because Im just happy with the learning I get. But after this paper was accepted, in the next, I would say, many years, there are more and more people realize, hey, before we actually, you know, do bug-finding things, lets first do a study, right, to understand, and then this paper was yeah I was very happy that it was cited many, many times.HUIZINGA: Yeah. And then gets the most influential paper many years later.LU: Many years later. Yes.HUIZINGA: Yeah, I feel like theres a lot of things going through my head right now, one of which is what AI is, is a pattern detector, and you were doing that before AI even came on the scene. Which goes to show you that humans are pretty good at pattern detection also. We might not do as fast as LU: True.HUIZINGA: as an AI but so this idea of learning from mistakes is a broad theme. Another theme that I see coming through your papers and your work is persistence. [LAUGHTER] And you mentioned this about your team, right. I was like, these people are people who dont give up. So we covered this idea in an Abstracts podcast recently talking about a paper which really brings this to light: If at First You Dont Succeed, Try, Try Again. Thats the name of the paper. And we didnt have time to discuss it in depth at the time because the Abstracts show is so quick. But we do now. So Id like you to expand a little bit on this big idea of persistence and how large language models are not only changing the way programming and verification happens but also providing insights into detecting retry bugs.LU: Yes. So I guess maybe I will, since you mentioned this persistence, you know, after that Learning from Mistakes paperso that was in 2008and in the next 10 years, a little bit more than 10 years, in terms of persistence, right, so we have continued, me and my students, my collaborators, we have continued working on, you know, finding concurrency bugs HUIZINGA: Yeah.LU: which is related to, kind of related to, why Im here at Microsoft Research. And we keep doing it, doing it, and then I feel like a high point was that I had a collaboration with my now colleagues here, Madan Musuvathi and Suman Nath. So we built a tool to detect concurrency bugs, and after more than 15 years of effort on this, we were able to find more than 1,000 concurrency bugs. It was built in a tool called Torch that was deployed in the company, and it won the Best Paper Award at the top system conference, SOSP, and it was actually a bittersweet moment. This paper seems to, you know, put an end HUIZINGA: Oh, interesting!LU: to our research. And also some of the findings from that paper is that we used to do very sophisticated program analysis to reason about the timing. And in that paper, we realized actually, sometimes, if youre a little bit fuzzy, dont aim to do perfect analysis, the resulting tool is actually more effective. So after that paper, Madan, Suman, and me, we kind of, you know, shifted our focus to looking at other types of bugs. And at the same time, the three of us realized the traditional, very precise program analysis may not be needed for some of the bug finding. So then, for this paper, this retry bugs, after we shifted our focus away from concurrency bugs, we realized, oh, there are many other types of important bugs, such as, in this case, like retry, right, when your software goes wrong, right. Another thing we learned is that it looks like you can never eliminate all bugs, so something will go wrong, [LAUGHTER] and then so thats why you need something like retry, right. So like if something goes wrong, at least you wont give up immediately.HUIZINGA: Right.LU: The software will retry. And another thing that started from this earlier effort is we started using large language models because we realized, yeah, you know, traditional program analysis sometimes can give you a very strong guarantee, but in some other cases, like in this retry case, some kind of fuzzy analysis, you know, not so precise, offered by large language models is sometimes even more beneficial. Yeah. So thats kind of, you know, the story behind this paper.HUIZINGA: Yeah, yeah, yeah, yeah. So, Shan, were hearing a lot about how large language models are writing code nowadays. In fact, NVIDIAs CEO says, mamas, dont let your babies grow up to be coders because AIs going to do that. I dont know if hes right, but one of the projects youre most excited about right now is called Verus, and your colleague Jay Lorch recently said that he sees a lot of synergy between AI and verification, where each discipline brings something to the other, and Rafah Hosn has referred to this as co-innovation or bidirectional enrichment. I dont know if thats exactly what is going on here, but it seems like it is. Tell us more about this project, Verus, and how AI and software verification are helping each other out.LU: Yes, yes, yes, yes. Im very excited about this project now! So first of all, starting from Verus. So Verus is a tool that helps you verify the correctness of Rust code. So this is a its a relatively new tool, but its creating a lot of, you know, excitement in the research community, and its created by my colleague Chris Hawblitzel and his collaborators outside Microsoft Research.HUIZINGA: Interesting.LU: And as I mentioned, right, this is a part that, you know, really inspired me. So traditionally to verify, right, your program is correct, it requires a lot of expertise. You actually have to write your proof typically in a special language. And, you know, so a lot of people, including me, right, who are so eager to get rid of bugs in my software, but there are people told me, saying just to learn that languageso they were referring to a language called Coqjust to learn that language, they said it takes one or two years. And then once you learn that language, right, then you have to learn about how to write proofs in that special language. So people, particularly in the bug-finding community, people know that, oh, in theory, you can verify it, but in reality, people dont do that. OK, so now going back to this Verus tool, why its exciting so it actually allows people to write proofs in Rust. So Rust is an increasingly popular language. And there are more and more people picking up Rust. Its the first time I heard about, oh, you can, you know, write proofs in a popular language. And also, another thing is in the past, you cannot verify an implementation directly. You can only verify something written in a special language. And the proof is proving something that is in a special language. And then finally, that special language is maybe then transformed into an implementation. So its just, theres just too many special languages there.HUIZINGA: A lot of layers.LU: A lot of layers. So now this Verus tool allows you to write a proof in Rust to prove an implementation that is in Rust. So its very direct. I just feel like Im just not good at learning a new language.HUIZINGA: Interesting.LU: So when I came here, you know, and learned about this Verus tool, you know, by Chris and his collaborators, I feel like, oh, looks like maybe I can give it a try. And surprisingly, I realized, oh, wow! I can actually write proofs using this Verus tool.HUIZINGA: Right.LU: And then, of course, you know, I was told, if you really want to, right, write proofs for large systems, it still takes a lot of effort. And then this idea came to me that, hey, maybe, you know, these days, like, large language models can write code, then why not let large language models write proofs, right? And of course, you know, other people actually had this idea, as well, but theres a doubt that, you know, can large language models really write proofs, right? And also, people have this feeling that, you know, large language models seem not very disciplined, you know, by nature. But, you know, thats what intrigued me, right. And also, I used to be a doubter for, say, GitHub Copilot. USED to! Because I feel like, yes, it can generate a lot of code, but who knows [LAUGHS] HUIZINGA: Whether its right LU: What, what is whether its right?HUIZINGA: Yeah.LU: Right, so I feel like, wow, you know, this could be a game-changer, right? Like, if AI can write not only code but also proofs. Yeah, so thats what I have been doing. Ive been working on this for one year, and I gradually get more collaborators both, you know, people in Microsoft Research Asia, and, you know, expertise here, like Chris, and Jay Lorch. They all help me a lot. So we actually have made a lot of progress.HUIZINGA: Yeah.LU: Like, now its, like, weve tried, like, for example, for some small programs, benchmarks, and we see that actually large language models can correctly prove the majority of the benchmarks that we throw to it. Yeah. Its very, very exciting.HUIZINGA: Well, and so and were going to talk a little bit more about some of those doubts and some of those interesting concerns in a bit. I do want you to address what I think Jay was getting at, which is that somehow the two help each other. The verification improves the AI. The AI improves the verification.LU: Yes, yes.HUIZINGA: How?LU: Yes. My feeling is that a lot of people, if theyre concerned with using AI, its because they feel like theres no guarantee for the content generated by AI, right. And then we also all heard about, you know, hallucination. And I tried myself. Like, I remember, at some point, if I ask AI, say, you know, which is bigger: is it three times three or eight? And the AI will tell me eight is bigger. And [LAUGHTER]HUIZINGA: Like, what?LU: So I feel like verification can really help AI HUIZINGA: Get better LU: because now you can give, you know, kind of, add in mathematical rigors into whatever that is generated by AI, right. And I say it would help AI. It will also help people who use AI, right, so that they know what can be trusted, right.HUIZINGA: Right.LU: What is guaranteed by this content generated by AI?HUIZINGA: Yeah, yeah, yeah.LU: Yeah, and now of course AI can help verification because, you know, verification, you know, its hard. There is a lot of mathematical reasoning behind it. [LAUGHS] And so now with AI, it will enable verification to be picked up by more and more developers so that we can get higher-quality software.HUIZINGA: Yeah.LU: Yeah.HUIZINGA: Yeah. And well get to that, too, about what I would call the democratization of things. But before that, I want to, again, say an observation that I had based on your work and my conversations with you is that youve basically dedicated your career to hunting bugs.LU: Yes.HUIZINGA: And maybe thats partly due to a personal story about how a tiny mistake became a bug that haunted you for years. Tell us the story.LU: Yes.HUIZINGA: And explain why and how it launched a lifelong quest to understand, detect, and expose bugs of all kinds.LU: Yes. So before I came here, I already had multiple times, you know, interacting with Microsoft Research. So I was a summer intern at Microsoft Research Redmond almost 20 years ago.HUIZINGA: Oh, wow!LU: I think it was in the summer of 2005. And I remember I came here, you know, full of ambition. And I thought, OK, you know, I will implement some smart algorithm. I will deliver some useful tools. So at that time, I had just finished two years of my PhD, so I, kind of, just started my research on bug finding and so on. And I remember I came here, and I was told that I need to program in C#. And, you know, I just naturally have a fear of learning a new language. But anyway, I remember, I thought, oh, the task I was assigned was very straightforward. And I think I went ahead of myself. I was thinking, oh, I want to quickly finish this, and I want to do something more novel, you know, that can be more creative. But then this simple task I was assigned, I ended up spending the whole summer on it. So the tool that I wrote was supposed to process very huge logs. And then the problem is my software is, like, you run it initially So, like, I can only run it for 10 minutes because my software used so much memory and it will crash. And then, I spent a lot of time I was thinking, oh, my software is just using too much memory. Let me optimize it, right. And then so, I, you know, I try to make sure to use memory in a very efficient way, but then as a result, instead of crashing every 10 minutes, it will just crash after one hour. And I know theres a bug at that time. So theres a type of bug called memory leak. I know theres a bug in my code, and I spent a lot of time and there was an engineer helping me checking my code. We spent a lot of time. We were just not able to find that bug. And at the end, we the solution is I was just sitting in front of my computer waiting for my program to crash and restart. [LAUGHTER] And at that time, because there was very little remote working option, so in order to finish processing all those logs, its like, you know, after dinner, I HUIZINGA: You have to stay all night!LU: I have to stay all night! And all my intern friends, they were saying, oh, Shan, you work really hard! And Im just feeling like, you know what Im doing is just sitting in front of my computer waiting [LAUGHTER]for my program to crash so that I can restart it! And near the end of my internship, I finally find the bug. It turns out that I missed a pair of brackets in one line of code.HUIZINGA: Thats it.LU: Thats it.HUIZINGA: Oh, my goodness.LU: And it turns out, because I was used to C, and in C, when you want to free, which means deallocate, an array, you just say free array. And if I remember correctly, in this language, C#, you have to say, free this array name and you put a bracket behind it. Otherwise, it will only free the first element. And I it was a nightmare. And I also felt like, the most frustrating thing is, if its a clever bug, right [LAUGHS]HUIZINGA: Sure.LU: then you feel like at least Im defeated by something complicated HUIZINGA: Smart.LU: Something smart. And then its like, you know, also all this ambition I had about, you know, doing creative work, right, with all these smart researchers in MSR (Microsoft Research), I feel like I ended up achieving very little in my summer internship.HUIZINGA: But maybe the humility of making a stupid mistake is the kind of thing that somebody whos good at hunting bugs Its like missing an error in the headline of an article, because the print is so big [LAUGHTER] that youre looking for the little things in the I know thats a journalists problem. Actually, I actually love that story. And it, kind of, presents a big picture of you, Shan, as a person who has a realistic, self-awareness of and humility, which I think is rare at times in the software world. So thanks for sharing that. So moving on. When we talked before, you mentioned the large variety of programming languages and how that can be a barrier to entry or at least a big hurdle to overcome in software programming and verification. But you also talked about, as we just mentioned, how LLMs have been a democratizing force LU: Yes.HUIZINGA: LU: Yes.HUIZINGA: and what you see now with the advent of tools like GitHub Copilot, LU: Yes.HUIZINGA: what whats changed?LU: Oh, so much has changed. Well, I dont even know how to start. Like, I used to be really scared about programming. You know, when I tell this story, a lot of people say, no, I dont believe you. And I feel like its a trauma, you know.HUIZINGA: Sure.LU: I almost feel like its like, you know, the college-day me, right, who was scared of starting any programming project. Somehow, I felt humiliated when asking those very, I feel like, stupid questions to my classmates. It almost changed my personality! Its like for a long time, whenever someone introduced me to a new software tool, my first reaction is, uh, I probably will not be able to successfully even install it. Like whenever, you know, theres a new language, my first reaction is, uh, no, Im not good at it. And then, like, for example, this GitHub Copilot thing, actually, I did not try it until I joined Microsoft. And then I, actually, I havent programmed for a long time. And then I started collaborating with people in Microsoft Research Asia, and he writes programs in Python, right. And I have never written a single line of Python code before. And also, this Verus tool. It helps you to verify code in Rust, but I have never learned Rust before. So I thought, OK, maybe let me just try GitHub Copilot. And wow! You know, its like I realized, wow! Like [LAUGHS]HUIZINGA: I can do this!LU: I can do this! And, of course, sometimes I feel like my colleagues may sometimes be surprised because on one hand it looks like Im able to just finish, you know, write a Rust function. But on some other days, I ask very basic questions, [LAUGHTER] and I have those questions because, you know, the GitHub Copilot just helps me finish! [LAUGHS]HUIZINGA: Right.LU: You know, Im just starting something to start it, and then it just helps me finish. And I wish, when I started my college, if at that time there was GitHub Copilot, I feel like, you know, my mindset towards programming and towards computer science might be different. So it does make me feel very positive, you know, about, you know, what future we have, you know, with AI, with computer science.HUIZINGA: OK, usually, I ask researchers at this time, what could possibly go wrong if you got everything right? And I was thinking about this question in a different way until just this minute. I want to ask you what do you think that it means to have a tool that can do things for you that you dont have to struggle with? And maybe, is there anything good about the struggle? Because youre framing it as it sapped your confidence.LU: [LAUGHS] Yes.HUIZINGA: And at the same time, I see a woman who emerged stronger because of this struggle with an amazing career, a huge list of publications, influential papers, citations, leadership role. [LAUGHTER] So in light of that LU: Right.HUIZINGA: what do you see as the tension between struggling to learn a new language versus having this tool that can just do it that makes you look amazing? And maybe the truth of it is you dont know!LU: Yeah. Thats a very good point. I guess you need some kind of balance. And on one hand, yes, I feel like, again, right, this goes back to like my internship. I left with the frustration that I felt like I have so much creativity to contribute, and yet I could not because of this language barrier. You know, I feel positive in the sense that just from GitHub Copilot, right, how it has enabled me to just bravely try something new. I feel like this goes beyond just computer science, right. I can imagine itll help people to truly unleash their creativity, not being bothered by some challenges in learning the tool. But on the other hand, you made a very good point. My adviser told me she feels like, you know, I write code slowly, but I tend to make fewer mistakes. And the difficulty of learning, right, and all these nightmares I had definitely made me more more cautious? I pay more respect to the task that is given to me, so there is definitely the other side of AI, right, which is, you feel like everything is easy and maybe you do not have the experience of those bugs, right, that a software can bring to you and you have overreliance, right, on this tool.HUIZINGA: Yeah!LU: So hopefully, you know, some of the things we were doing now, right, like for example, say verification, right, like bringing this mathematical rigor to AI, hopefully that can help.HUIZINGA: Yeah. You know, even as you unpack the nuances there, it strikes me that both are good. Both having to struggle and learning languages and understanding LU: Yeah.HUIZINGA: the core of it and the idea that in natural language, you could just say, heres what I want to happen, and the AI does the code, the verification, etc. That said, do we trust it? And this was where I was going with the first what could possibly go wrong? question. How do we know that it is really as clever as it appears to be? [LAUGHS]LU: Yeah, I think I would just use the research problem we are working on now, right. Like, I think on one hand, I can use AI to generate a proof, right, to prove the code generated by AI is correct. But having said that, even if were wildly successful, you know, in this thing, human beings expertise is still needed because just take this as an example. What do you mean by correct, right?HUIZINGA: Sure.LU: And so someone first has to define what correctness means. And then so far, the experience shows that you cant just define it using natural language because our natural language is inherently imprecise.HUIZINGA: Sure.LU: So you still need to translate it to a formal specification in a programming language. It could be in a popular language like in Rust, right, which is what Verus is aiming at. And then we are, like, for example, some of the research we do is showing that, yes, you know, I can also use AI to do this translation from natural language to specification. But again, then, who to verify that, right? So at the end of the day, I think we still do need to have humans in the loop. But what we can do is to lower the burden and make the interface not so complicated, right. So that itll be easy for human beings to check what AI has been doing.HUIZINGA: Yeah. You know, everything were talking about just reinforces this idea that were living in a time where the advances in computer science that seemed unrealistic or impossible, unattainable even a few years ago are now so common that we take it for granted. And they dont even seem outrageous, but they are. So Im interested to know what, if anything, you would classify now as blue sky research in your field. Maybe something in systems research today that looks like a moonshot. Youve actually anchored this in the fact that you, kind of, have, you know, blinders on for the work youre doinghead down in the in the work youre doingbut even as you peek up from the work that might be outrageous, is there anything else? I just like to get this out there that, you know, whats going on 10 years down the line?LU: You know, sometimes I feel like Im just now so much into my own work, but, you know, occasionally, like, say, when I had a chat with my daughter and I explained to her, you know, oh, Im working on, you know, not only having AI to generate code but also having AI to prove, right, the code is correct. And she would feel, wow, that sounds amazing! [LAUGHS] So I dont know whether that is, you know, a moonshot thing, but thats a thing that Im super excited about HUIZINGA: Yeah.LU: about the potential. And then there also have, you know, my colleagues, we spend a lot of time building systems, and its not just about correctness, right. Like, the verification thing Im doing now is related to automatically verify its correct. But also, you need to do a lot of performance tuning, right. Just so that your system can react fast, right. It can have good utilization of computer resources. And my colleagues are also working on using AI, right, to automatically do performance tuning. And I know what they are doing, so I dont particularly feel thats a moonshot, but I guess HUIZINGA: I feel like, because you are so immersed, [LAUGHTER] that you just dont see how much we think LU: Yeah!HUIZINGA: its amazing. Well, Im just delighted to talk to you today, Shan. As we close and youve sort of just done a little vision casting, but lets take your daughter, my daughter,[LAUGHTER] all of our daughters LU: Yes!HUIZINGA: How does what we believe about the future in terms of these things that we could accomplish influence the work we do today as sort of a vision casting for the next Shan Lu whos struggling in undergrad/grad school?LU: Yes, yes, yes. Oh, thank you for asking that question. Yeah, I have to say, you know, I think were in a very interesting time, right, with all this AI thing.HUIZINGA: Isnt that a curse in China? May you live in interesting times!LU: And I think there were times, actually, you know, before I myself fully embraced AI, I was indeed I had my daughter in mind. I was worried when she grows up, what would happen? There will be no job for her because everything will be done by AI!HUIZINGA: Oh, interesting.LU: But then now, now that I have, you know, kind of fully embraced AI myself, actually, I see this more and more positive. Like you said, I remember, you know, those older days myself, right. That is really, like, I have this struggle that I feel like I can do better. I feel like I have ideas to contribute, but just for whatever reason, right, it took me forever to learn something which I feel like its a very mechanical thing, but it just takes me forever to learn, right. And then now actually, I see this hope, right, with AI, you know, a lot of mechanical things that can actually now be done in a much more automated way by AI, right. So then now truly, you know, my daughter, many girls, many kids out there, right, whatever you know, they are good at, their creativity, itll be much easier, right, for them to contribute their creativity to whatever discipline they are passionate about. Hopefully, they dont have to, you know, go through what I went through, right, to finally be able to contribute. But then, of course, you know, at the same time, I do feel this responsibility of me, my colleagues, MSR, we have the capability and also the responsibility, right, of building AI tools in a responsible way so that it will be used in a positive way by the next generation.HUIZINGA: Yeah. Shan Lu, thank you so much for coming on the show today. [MUSIC] Its been absolutely delightful, instructive, informative, wonderful.LU: Thank you. My pleasure.

In this edition:We introduce privacy enhancements for multiparty deep learning, a framework using smaller, open-source models to provide relevance judgments, and other notable new research.We congratulate Yasuyuki Matsushita, who was named an IEEE Computer Society Fellow.Weve included a recap of the extraordinary, far-reaching work done by researchers at Microsoft in 2024.NEW RESEARCHAI meets materials discoveryTwo of the transformative tools that play a central role in Microsofts work on AI for science are MatterGen and MatterSim. In the world of materials discovery, each plays a distinct yet complementary role in reshaping how researchers design and validate new materials.Read the storyNEW RESEARCHDistributed training enables multiple parties to jointly train a machine learning model on their respective datasets, which can help address the challenges posed by requirements in modern machine learning for large volumes of diverse data. However, this can raise security and privacy issues protecting each partys data during training and preventing leakage of private information from the model after training through various inference attacks.In a recent paper, Communication Efficient Secure and Private Multi-Party Deep Learning, researchers from Microsoft address these concerns simultaneously by designing efficient Differentially Private, secure Multiparty Computation (DP-MPC) protocols for jointly training a model on data distributed among multiple parties. This DP-MPC protocol in the two-party setting is 56-to-794 times more communication-efficient and 16-to-182 times faster than previous such protocols. This work simplifies and improves on previous attempts to combine techniques from secure multiparty computation and differential privacy, especially in the context of training machine learning models.Read the paperNEW RESEARCHTraining and evaluating retrieval systems requires significant relevance judgments, which are traditionally collected from human assessors. This process is both costly and time-consuming. Large language models (LLMs) have shown promise in generating relevance labels for search tasks, offering a potential alternative to manual assessments. Current approaches often rely on a single LLM. While effective, this approach can be expensive and prone to intra-model biases that can favor systems leveraging similar models.In a recent paper: JudgeBlender: Ensembling Judgments for Automatic Relevance Assessment, researchers from Microsoft we introduce a framework that employs smaller, open-source models to provide relevance judgments by combining evaluations across multiple LLMs (LLMBlender) or multiple prompts (PromptBlender). By leveraging the LLMJudge benchmark, they compare JudgeBlender with state-of-the-art methods and the top performers in the LLMJudge challenge. This research shows that JudgeBlender achieves competitive performance, demonstrating that very large models are often unnecessary for reliable relevance assessments.Read the paperNEW RESEARCHCongestion games are used to describe the behavior of agents who share a set of resources. Each player chooses a combination of resources, which may become congested, decreasing utility for the players who choose them. Players can avoid congestion by choosing combinations that are less popular. This is useful for modeling a range of real-world scenarios, such as traffic flow, data routing, and wireless communication networks.In a recent paper: Convergence to Equilibrium of No-regret Dynamics in Congestion Games; researchers from Microsoft and external colleagues propose CongestEXP, a decentralized algorithm based on the classic exponential weights method. They evaluate CongestEXP in a traffic congestion game setting. As more drivers use a particular route, congestion increases, leading to higher travel times and lower utility. Players can choose a different route every day to optimize their utility, but the observed utility by each player may be subject to randomness due to uncertainty (e.g., bad weather). The researchers show that this approach provides both regret guarantees and convergence to Nash Equilibrium, where no player can unilaterally improve their outcome by changing their strategy.Read the paperNEW RESEARCHResearch and development (R&D) plays a pivotal role in boosting industrial productivity. However, the rapid advance of AI has exposed the limitations of traditional R&D automation. Current methods often lack the intelligence needed to support innovative research and complex development tasks, underperforming human experts with deep knowledge.LLMs trained on vast datasets spanning many subjects are equipped with extensive knowledge and reasoning capabilities that support complex decision-making in diverse workflows. By autonomously performing tasks and analyzing data, LLMs can significantly increase the efficiency and precision of R&D processes.In a recent article, researchers from Microsoft introduce RD-Agent, a tool that integrates data-driven R&D systems and harnesses advanced AI to automate innovation and development.At the heart of RD-Agent is an autonomous agent framework with two key components: a) Research and b) Development. Research focuses on actively exploring and generating new ideas, while Development implements these ideas. Both components improve through an iterative process, illustrated in Figure 1 of the article, ensures the system becomes increasingly effective over time.Read the articleSpotlight: Microsoft research newsletterMicrosoft Research NewsletterStay connected to the research community at Microsoft.Subscribe todayOpens in a new tab Microsoft Research | In case you missed itMicrosoft Research 2024: A year in reviewDecember 20, 2024Microsoft Research did extraordinary work this year, using AI and scientific research to make progress on real-world challenges like climate change, food security, global health, and human trafficking. Heres a look back at the broad range of accomplishments and advances in 2024. AIOpsLab: Building AI agents for autonomous cloudsDecember 20, 2024AIOpsLab is a holistic evaluation framework for researchers and developers, to enable the design, development, evaluation, and enhancement of AIOps agents, which also serves the purpose of reproducible, standardized, interoperable, and scalable benchmarks. Yasuyuki Matsushita, IEEE Computer Society 2025 FellowDecember 19, 2024Congratulations to Yasuyuki Matsushita, Senior Principal Research Manager at Microsoft Research, who was named a 2025 IEEE Computer Society Fellow. Matsushita was recognized for contributions to photometric 3D modeling and computational photography. View more news and awards Opens in a new tab

Materials innovation is one of the key drivers of major technological breakthroughs. The discovery of lithium cobalt oxide in the 1980s laid the groundwork for todays lithium-ion battery technology. It now powers modern mobile phones and electric cars, impacting the daily lives of billions of people. Materials innovation is also required for designing more efficient solar cells, cheaper batteries for grid-level energy storage, and adsorbents to recycle CO2 from atmosphere.Finding a new material for a target application is like finding a needle in a haystack. Historically, this task has been done via expensive and time-consuming experimental trial-and-error. More recently, computational screening of large materials databases has allowed researchers to speed up this process. Nonetheless, finding the few materials with the desired properties still requires the screening of millions of candidates.Today, in a paper published in Nature (opens in new tab), we share MatterGen, a generative AI tool that tackles materials discovery from a different angle. Instead of screening the candidates, it directly generates novel materials given prompts of the design requirements for an application. It can generate materials with desired chemistry, mechanical, electronic, or magnetic properties, as well as combinations of different constraints. MatterGen enables a new paradigm of generative AI-assistedmaterials design that allows for efficient exploration of materials, going beyond the limited set of known ones.Figure 1: Schematic representation of screening and generative approaches to materials designA novel diffusion architectureMatterGen is a diffusion model that operates on the 3D geometry of materials. Much like an image diffusion model generates pictures from a text prompt by modifying the color of pixels from a noisy image, MatterGen generates proposed structures by adjusting the positions, elements, and periodic lattice from a random structure. The diffusion architecture is specifically designed for materials to handle specialties like periodicity and 3D geometry.Figure 2: Schematic representation of MatterGen: a diffusion model to generate novel and stable materials. MatterGen can be fine-tuned to generate materials under different design requirements such as specific chemistry, crystal symmetry, or materials properties.The base model of MatterGen achieves state-of-the-art performance in generating novel, stable, diverse materials (Figure 3). It is trained on 608,000 stable materials from the Materials Project (opens in new tab) (MP) and Alexandria (opens in new tab) (Alex) databases. The performance improvement can be attributed to both the architecture advancements, as well asthe quality and size of our training data.Figure 3: Performance of MatterGen and other methods in the generation of stable, unique, and novel structures. The training dataset for each method is indicated in parentheses. The purple bar highlights performance improvements due to MatterGens architecture alone, while the teal bar highlights performance improvements that come also from the larger training dataset.MatterGen can be fine-tuned with a labelled dataset to generate novel materials given any desired conditions. We demonstrate examples of generating novel materials given a targets chemistry and symmetry, as well as electronic, magnetic, and mechanical property constraints (Figure 2). Outperforming screeningFigure 4: Performance of MatterGen (teal) and traditional screening (yellow) in finding novel, stable, and unique structures that satisfy the design requirement of having bulk modulus greater than 400 GPa.The key advantage of MatterGen over screening is its ability to access the full space of unknown materials. In Figure 4, we show that MatterGen continues to generate more novel candidate materials with high bulk modulus above 400 GPa, for example, which are hard to compress. In contrast, screening baseline saturates due to exhausting known candidates.Spotlight: Blog postMedFuzz: Exploring the robustness of LLMs on medical challenge problemsMedfuzz tests LLMs by breaking benchmark assumptions, exposing vulnerabilities to bolster real-world accuracy.Read moreOpens in a new tab Handling compositional disorderFigure 5: Illustration of compositional disorder. Left: a perfect crystal without compositional disorder and with a repeating unit cell (black dashed). Right: crystal with compositional disorder, where each site has 50% probability of yellow and teal atoms.Compositional disorder (Figure 5) is a commonly observed phenomenon where different atoms can randomly swap their crystallographic sites in a synthesized material. Recently (opens in new tab), the community has been exploring what it means for a material to be novel in the context of computationally designed materials, as widely employed algorithms will not distinguish between pairs of structures where the only difference is a permutation of similar elements in their respective sites.We provide an initial solution to this issue by introducing a new structure matching algorithm that considers compositional disorder. The algorithm assesses whether a pair of structures can be identified as ordered approximations of the same underlying compositionally disordered structure. This provides a new definition of novelty and uniqueness, which we adopt in our computational evaluation metrics. We also make our algorithm publicly available (opens in new tab) as part of our evaluation package.Experimental lab verificationFigure 6: Experimental validation of the proposed compound, TaCr2O6In addition to our extensive computational evaluation, we have validated MatterGens capabilities through experimental synthesis. In collaboration with the team led by Prof Li Wenjie from the Shenzhen Institutes of Advanced Technology (opens in new tab) (SIAT) of the Chinese Academy of Sciences, we have synthesized a novel material, TaCr2O6, whose structure was generated by MatterGen after conditioning the model on a bulk modulus value of 200 GPa. The synthesized materials structure aligns with the one proposed by MatterGen, with the caveat of compositional disorder between Ta and Cr. Additionally, we experimentally measure a bulk modulus of 169 GPa against the 200 GPa given as design specification, with a relative error below 20%, very close from an experimental perspective. If similar results can be translated to other domains, it will have a profound impact on the design of batteries, fuel cells, and more. AI emulator and generator flywheelMatterGen presents a new opportunity for AI accelerated materials design, complementing our AI emulator MatterSim. MatterSim follows the fifth paradigm of scientific discovery, significantly accelerating the speed of material properties simulations. MatterGen in turn accelerates the speed of exploring new material candidates with property guided generation. MatterGen and MatterSim can work together as a flywheel to speed up both the simulation and exploration of novel materials.Making MatterGen availableWe believe the best way to make an impact in materials design is to make our model available to the public. We release the source code of MatterGen (opens in new tab) under the MIT license, together with the training and fine-tuning data. We welcome the community to use and build on top of our model.MatterGen represents a new paradigm of materials design enabled by generative AI technology. It explores a significantly larger space of materials than screening-based methods. It is also more efficient by guiding materials exploration with prompts. Similar to how generative AI has impacted drug discovery (opens in new tab), it will have profound impact on how we design materials in broad domains including batteries, magnets, and fuel cells.We plan to continue our work with external collaborators to further develop and validate the technology. At the Johns Hopkins University Applied Physics Laboratory (APL), were dedicated to the exploration of tools with the potential to advance discovery of novel, mission-enabling materials. Thats why we are interested in understanding the impact that MatterGen could have on materials discovery, said Christopher Stiles, a computational materials scientists leading multiple materials discovery efforts at APL.AcknowledgementThis work is the result of highly collaborative team efforts at Microsoft Research AI for Science. The full authors include: Claudio Zeni, Robert Pinsler, Daniel Zgner, Andrew Fowler, Matthew Horton, Xiang Fu, Zilong Wang, Aliaksandra Shysheya, Jonathan Crabb, Shoko Ueda, Roberto Sordillo, Lixin Sun, Jake Smith, Bichlien Nguyen, Hannes Schulz, Sarah Lewis, Chin-Wei Huang, Ziheng Lu, Yichi Zhou, Han Yang, Hongxia Hao, Jielan Li, Chunlei Yang, Wenjie Li, Ryota Tomioka, Tian Xie.Opens in a new tab

Transcript[TEASER][MUSIC PLAYS UNDER DIALOGUE]TIAN XIE: Yeah,ZIHENG LU: Previously, a lot of people are using this atomistic simulator and this generative models alone. But if you think about it, now that we have these two foundation models together, it really can make things different, right. You have a very good idea generator. And you have a very good goalkeeper. And you put them together. They form a loop. And now you can use this loop to design materials really quickly.[TEASER ENDS]LINDSAY KALTER: Youre listening to Ideas, a Microsoft Research Podcast that dives deep into the world of technology research and the profound questions behind the code. In this series, well explore the technologies that are shaping our future and the big ideas that propel them forward.[MUSIC FADES]Im your guest host, Lindsay Kalter. Today Im talking to Microsoft Principal Research Manager Tian Xie and Microsoft Principal Researcher Ziheng Lu. Tian is doing fascinating work with MatterGen, an AI tool for generating new materials guided by specific design requirements. Ziheng is one of the visionaries behind MatterSim, which puts those new materials to the test through advanced simulations. Together, theyre redefining whats possible in materials science. Tian and Ziheng, welcome to the podcast.TIAN XIE: Very excited to be here.ZIHENG LU: Thanks, Lindsay, very excited.KALTER: Before we dig into the specifics of MatterGen and MatterSim, lets give our audience a sense of how you, as researchers, arrived at this moment. Materials science, especially at the intersection of computer science, is such a cutting-edge and transformative field. What first drew each of you to this space? And what, if any, moment or experience made you realize this was where you wanted to innovate? Tian, do you want to start?XIE: So I started working on AI for materials back in 2015, when I started my PhD. So I come as a chemist and materials scientist, but I was, kind of, figuring out what I want to do during my PhD. So there is actually one moment really drove me into the field. That was AlphaGo. AlphaGo was, kind of, coming out in 2016, where it was able to beat the world champion in go in 2016. I was extremely impressed by that because I, kind of, learned how to do go, like, in my childhood. I know how hard it is and how much effort those professional go players have spent, right, in learning about go. So I, kind of, have the feeling that if AI can surpass the world-leading go players, one day, it will too surpass material scientists, right, in their ability to design novel materials. So thats why I ended up deciding toLU: Thats very interesting, Tian. So, actually, I think I started, like, two years before you as a PhD student. So I, actually, I was trained as a computational materials scientist solely, not really an AI expert. But at that time, the computational materials science did not really work that well. It works but not working that well. So after, like, two or three years, I went back to experiments for, like, another two or three years because, I mean, the experiment is always the gold standard, right. And I worked on this experiments for a few years, and then about three years ago, I went back to this field of computation, especially because of AI. At that time, I think GPT and these large AI models that currently were using is not there, but we already have their prior forms like BERT, so we see the very large potential of AI. We know that these large AIs might work. So one idea is really to use AI to learn the entire space of materials and really grasp the physics there, and that really drove me to this field and thats why Im here working on this field, yeah.KALTER: Were going to get into what MatterGen and MatterSim mean for materials sciencethe potential, the challenges, and open questions. But first, give us an overview of what each of these tools are, how they do what they do, andas this show is about big ideasthe idea driving the work. Ziheng, lets have you go first.LU: So MatterSim is a tool to do in silico characterizations of materials. If you think about working on materials, you have several steps. You first need to synthesize it, and then you need to characterize this. Basically, you need to know what property, what structures, whatever stuff about these materials. So for MatterSim, what we want to do is to really move the characterization process, a lot of these processes, into using computations. So the idea behind MatterSim is to really learn the fundamentals of physics. So we learn the energies and forces and stresses from these atomic structures and the charge densities, all of these things, and then with these, we can really simulate any sort of materials using our computational machines. And then with these, we can really characterize a lot of these materials properties using our computer, that is very fast. Its much faster than we do experiments so that we can accelerate the materials design. So just in a word, basically, you input your material into your computer, a structure into your computer, and MatterSim will try to simulate these materials like what you do in a furnace or with an XRDKALTER: All right, thank you very much. Tian, why dont you tell us about MatterGen?XIE: Yeah, thank you. So, actually, Ziheng, once you start with explaining MatterSim, it makes it much easier for me to explain MatterGen. So MatterGen actually represents a new way to design materials with generative AI. Material discovery is like finding needles in a haystack. Youre looking for a material with a very specific property for a material application. For example, like finding a room-temperature superconductor or finding a solid that can conduct a lithium ion very well inside a battery. So its like finding one very specific material from a million, kind of, candidates. So the conventional way of doing material discovery is via screening, where you, kind of, go over millions of candidates to find the one that youre looking for, where MatterSim is able to significantly accelerate that process by making the simulation much faster. But its still very inefficient because you need to go through this million candidates, right. So with MatterGen, you can, kind of, directly generate materials given the prompts of the design requirements for the application. So this means that you can discover materialsdiscover useful materials much more efficiently. And it also allows us to explore a much larger space beyond the set of known materials.KALTER: Thank you, Tian. Can you tell us a little bit about how MatterGen and MatterSim work together?XIE: So you can really think about MatterSim and MatterGen accelerating different parts of materials discovery process. MatterSim is trying to accelerate the simulation of material properties, while MatterGen is trying to accelerate the search of novel material candidates. It means that they can really work together as a flywheel and you can compound the acceleration from both models. They are also both foundation AI models, meaning they can both be used for a broad range of materials design problems. So were really looking forward to see how they can, kind of, working together iteratively as a tool to design novel materials for a broad range of applications.LU: I think thats a very good, like, general introduction of how they work together. I think I can provide an example of how they really fit together. If you want a material with a specific, like, bulk modulus or lithium-ion conductivity or thermal conductivity for your CPU chips, so basically what you want to do is start with a pool of material structures, like some structures from the database, and then you compute or you characterize your wanted property from that stack of materials. And then what you do, youve got these properties and structure pairs, and you input these pairs into MatterGen. And MatterGen will be able to give you a lot more of these structures that are highly possible to be real. But the number will be very large. For example, for the bulk modulus, I dont remember the number we generated in our work was that like thousands, tens of thousands?XIE: Thousands, tens of thousands.LU: Yeah, that would be a very large number pool even with MatterGen, so then the next step will be, how would you like to screen that? You cannot really just send all of those structures to a lab to synthesize. Its too much, right. Thats when MatterSim again comes in. So MatterSim comes in and screen all those structures again and see which ones are the most likely to be synthesized and which ones have the closest property you wanted. And then after screening, you probably get five, 10 top candidates and then you send to a lab. Boom, everything goes down. Thats it.KALTER: Im wondering if theres any prior research or advancements that you drew from in creating MatterGen and MatterSim. Were there any specific breakthroughs that influenced your approaches at all?LU: Thanks, Lindsay. I think Ill take that question first. So interestingly for MatterSim, a very fundamental idea was drew from Chi Chen, who was a previous lab mate of mine and now also works for Microsoft at Microsoft Quantum. He made this fantastic model named M3GNet, which is a prior form of a lot of these large-scale models for atomistic simulations. That model, M3GNet, actually resolves the near ground state prediction problem. I mean, the near ground state problem sounds like a fancy but not realistic word, but what that actually means is that it can simulate materials at near-zero covalent states. So basically at very low temperatures. So at that time, we were thinking since the models are now able to simulate materials at their near ground states, its not a very large space. But if you also look at other larger models, like GPT whatever, those models are large enough to simulate entire human language. So its possible to really extend the capability from these such prior models to very large space. Because we believe in the capability of AI, then it really drove us to use MatterSim to learn the entire space of materials. I mean, the entire space really means the entire periodic table, all the temperatures and the pressures people can actually grasp.XIE: Yeah, I still remember a lot of the amazing works from Chi Chen whenever were, kind of, back working on property-prediction models. So, yeah, so the problem of generating materials from properties is actually a pretty old one. I still remember back in 2018, when I was, kind of, working on CGCNN (crystal graph convolutional neural networks) and giving a talk about property-prediction models, right, one of the first questions people asked is, OK, can you inverse this process? Instead of going from material structure to properties, can you, kind of, inversely generate the materials directly from their property conditions? So in a way, this is, kind of, like a dream for material scientistssome people even call it, like, holy grailbecause, like, the end goal is really about finding materials property, right, [that] will satisfy your application. So Ive been, kind of, thinking about this problem for a while and also there has been a lot of work, right, over the past few years in the community to build a generative model for materials. A lot of people have tried before, like 2020, using ideas like VAEs or GANs. But its hard to represent materials in this type of generative model architecture, and many of those models generated relatively poor candidates. So I thought it was a hard problem. I, kind of, know it for a while. But there is no good solutions back then. So I started to focus more on this problem during my postdoc, when I studied that in 2020 and I keep working on that in 2021. At the beginning, I wasnt really sure exactly what approach to take because its, kind of, like open question and really tried a lot of random ideas. So one day actually in my group back then with Tommi Jaakkola and Regina Barzilay at MITs CSAIL (Computer Science & Artificial Intelligence Laboratory), we, kind of, get to know this method called diffusion model. It was a very early stage of a diffusion model back then, but it already began to show very promising signs, kind of, achieving state of art in many problems like 3D point cloud generation and the 3D molecular conformer generation. So the work that really inspired me a lot is two works that was for molecular conformer generation. One is ConfGF, and one is GeoDiff. So they, kind of, inspired me to, kind of, focus more on diffusion models. That actually lead to CDVAE (crystal diffusion variational autoencoder). So its interesting that we, kind of, spend like a couple of weeks in trying all this diffusion idea, and without that much work, it actually worked quite out of box. And at that time, CDVAE achieves much better performance than any previous models in materials generation, and were, kind of, super happy with that. So after CDVAE, I, kind of, joined Microsoft, now working with more people together on this problem of generative model for materials. So we, kind of, know what the limitations of CDVAE are, is that it can do unconditional material generation well means it can generate novel material structures, but it is very hard to use CDVAE to do property-guided generations. So basically, it uses an architecture called a variational autoencoder, where you have a latent space. So the way that you do property-guided generation there was to do a, kind of, a gradient update inside the latent space. But because the latent space wasnt learned very well, so it actually you cannot do, kind of, good property-guided generation. We only managed to do energy-guided generation, but it wasnt successful in going beyond energy. So that comes us to really thinking, right, how can we make the property-guided generation much better? So I remember like one day, actually, my colleague, Daniel Zgner, who actually really showed me this blog which basically explains this idea of classifier-free guidance, which is the powerhouse behind the text-image generative models. And so, yeah, then we began to think about, can we actually make the diffusion model work for classifier-free guidance? That lead us to remove the, kind of, the variational autoencoder component from CDVAE and begin to work on a pure diffusion architecture. But then there was, kind of, a lot of development around that. But it turns out that classifier-free guidance is the key really to make property-guided generation work, and then combined with a lot more effort in, kind of, improving architecture and also generating more data and also trying out all these different downstream tasks that end up leading into MatterGen as we see today.KALTER: Yeah, I think youve both done a really great job of explaining how MatterGen and MatterSim work together and how MatterGen can offer a lot in terms of reducing the amount of time and work that goes into finding new materials. Tian, how does the process of using MatterGen to generate materials translate into real-world applications?XIE: Yeah, thats a fantastic question. So one way that I think about MatterGen, right, is that you can think about it as like a copilot for materials scientists, right. So they can help you to come up with, kind of, potential good hypothesis for the materials design problems that youre looking for. So say youre trying to design a battery, right. So you may have some ideas over, OK, what candidates you want to make, but this is, kind of, based on your own experience, right. Depths of experience as a researcher. But MatterGen is able to, kind of, learn from a very broad set of data, so therefore, it may be able to come up with some good suggestions, even surprising suggestions, for you so that you can, kind of, try this out, right, both with computation or even one day in wet lab and experimentally synthesize it. But I also want to note that this, in a way, this is still an early stage in generative AI for materials means that I dont expect all the candidates MatterGen generates will be, kind of, suits your needs, right. So you still need to, kind of, look into them with expertise or with some kind of computational screening. ButKALTER: I want to pivot a little bit to the MatterSim side of things. I know identifying new combinations of compounds is key to meeting changing needs for things like sustainable materials. But testing them is equally important to developing materials that can be put to use. Ziheng, how does MatterSim handle the uncertainty of how materials behave under various conditions, and how do you ensure that the predictions remain robust despite the inherent complexity of molecular systems?LU: Thanks. Thats a very, very good question. So uncertainty quantification is a key to make sure all these predictions and simulations are trustworthy. And thats actually one of the questions we got almost every time after a presentation. So people will ask, wellespecially those experimentalistswould ask, well, Ive been using your model; how do I know those predictions are true under the very complex conditions Im using in my experiments? So to understand how we deal with uncertainty, we need to know how MatterSim really functions in predicting an arbitrary property, especially under the condition you want, like the temperature and pressure. That would be quite complex, right? So in the ideal case, we would hope that by using MatterSim, you can directly simulate the properties you want using molecular dynamics combined with statistical mechanics. So if so, it would be easy to really quantify the uncertainty because there are just two parts: the error from the model and the error from the simulations, the statistical mechanics. So the error from the model will be able to be measured by, what we call, an ensemble. So basically you start with different random seeds when you train the model, and then when you predict your property, you use several models from the ensemble and then you get different numbers. If the variance from the numbers are very large, youll say the prediction is not that trustworthy. But a lot of times, we will see the variance is very small. So basically, an ensemble of several different models will give you almost exactly the same number; youre quite sure that the number is somehow very, like, useful. So thats one level of the way we want to get our property. But sometimes, its very hard to really directly simulate the property you want. For example, for catalytic processes, its very hard to imagine how you really get those coefficients. Its very hard. The process is just too complicated. So for that process, what we do is to really use the, what we call, embeddings learned from the entire material space. So basically that vector we learned for any arbitrary material. And then start from that, we build a very shallow layer of a neural network to predict the property, but that also means you need to bring in some of your experimental or simulation data from your side. And for that way of predicting a property to measure the uncertainty, its still like the two levels, right. So we dont really have the statistical error anymore, but what we have is, like, only the model error. So you can still stick to the ensemble, and then it will work, right. So to be short, so MatterSim can provide you an uncertainty to make sure the prediction tells you whether its true or not.KALTER: So in many ways, MatterSim is the realist in the equation, and its there to sort of be a gatekeeper for MatterGen, which is the idea generator.XIE: I really like the analogy.LU: Yeah.KALTER: As is the case with many AI models, the development of MatterGen and MatterSim relies on massive amounts of data. And here you use a simulation to create the needed training data. Can you talk about that process and why youve chosen that approach, Tian?XIE: So one advantage here is that we can really use large-scale simulation to generate data. So we have a lot of compute here at Microsoft on our Azure platform, right. So how we generate the data is that we use a method called density functional theory, DFT, which is a quantum mechanical method. And we use a simulation workflow built on top with DFT to simulate the stability of materials. So what we do is that we curate a huge amount of material structures from multiple different sources of open data, mostly including Materials Project and Alexandria database, and in total, there are around 3 million materials candidates coming from these two databases. But not all of these structures, they are stable. So therefore, we try to use DFT to compute their stability and try to filter down the candidates such that we are making sure that our training data only have the most stable ones. This leads into around 600,000 training data, which was used to train the base model of MatterGen. So I want to note that actually we also use MatterSim as part of the workflow because MatterSim can be used to prescreen unstable candidates so that we dont need to use DFT to compute all of them. I think at the end, we computed around 1 million DFT calculations where two-thirds of them, they are already filtered out by MatterSim, which saves us a lot of compute in generating our training data.LU: Tian, you have a very good description of how we really get those ground state structures for the MatterGen model. Actually, weve been also using MatterGen for MatterSim to really get the training data. So if you think about the simulation space of materials, its extremely large. So we would think it in a way that it has three axis, so basically the elements, the temperature, and the pressure. So if you think about existing databases, they have pretty good coverage of the elements space. Basically, we think about Materials Project, NOMAD, they really have this very good coverage of lithium oxide, lithium sulfide, hydrogen sulfide, whatever, those different ground-state structures. But they dont really tell you how these materials behave under certain temperature and pressure, especially under those extreme conditions like 1,600 Kelvin, which you really use to synthesize your materials. Thats where we really focused on to generate the data for MatterSim. So its really easy to think about how we generate the data, right. You put your wanted material into a pressure cooker, basically, molecular dynamics; it can simulate the materials behavior on the temperature and pressure. So thats it. Sounds easy, right? But thats not true because what we want is not one single material. What we want is the entire material space. So that will be making the effort almost impossible because the space is just so large. So thats where we really develop this active learning pipeline. So basically, what we do is, like, we generate a lot of these structures for different elements and temperatures, pressures. Really, really a lot. And then what we do is, like, we ask the active learning or the uncertainty measurements to really say whether the model knows about this structure already. So if the model thinks, well, I think I know the structure already. So then, we dont really calculate this structure using density function theory, as Tian just said. So this will really save us like 99% of the effort in generating the data. So in the end, by combining this molecular dynamics, basically pressure cooker, together with active learning, we gathered around 17 million data for MatterSim. So that was used to train the model. And now it can cover the entire periodic table and a lot of temperature and pressures.KALTER: Thank you, Ziheng. Now, Im sure this is not news to either one of you, given that youre both at the forefront of these efforts, but there are a growing number of tools aimed at advancing materials science. So what is it about MatterGen and MatterSim in their approach or capabilities that distinguish them?XIE: Yeah, I think I can start. So I think there is, in the past one year, there is a huge interest in building up generative AI tools for materials. So we have seen lots and lots of innovations from the community published in top conferences like NeurIPS, ICLR, ICML, etc. So I think what distinguishes MatterGen, in my point of view, are two things. First is that we are trained with a very big dataset that we curated very, very carefully, and we also spent quite a lot of time to refining our diffusion architecture, which means that our model is capable of generating very, kind of, high-quality, highly stable and novel materials. We have some kind of bar plot in our paper showcasing the advantage of our performance. I think thats one key aspect. And I think the second aspect, which in my point of view is even more important, is that it has the ability to do property-guided generation. Many of the works that we saw in the community, they are more focused on the problem of crystal structure prediction, which MatterGen can also do, but we focus more on really property-guided generation because we think this is one of the key problems that really materials scientists care about. So the ability to do a very broad range of property-guided generationand we have, kind of, both computational and now experimental result to validate thoseI think thats the second strong point for MatterGen.KALTER: Ziheng, do you want to add to that?LU: Yeah, thanks, Lindsay. So on the MatterSim side, I think its really the diverse condition it can handle that makes a difference. Weve been talking about, like, the training data we collected really covers the entire periodic table and also, more importantly, the temperatures from 0 Kelvin to 5,000 Kelvin and the pressures from 0 gigapascal to 1,000 gigapascal. That really covers what humans can control nowadays. I mean, its very hard to go beyond that. If you know anyone [who] can go beyond that, let me know. So that really makes MatterSim different. Like, it can handle the realistic conditions. I think beyond that, I would say the combo between MatterSim and MatterGen really makes these set of tools really different. So previously, a lot of people are using this atomistic simulator and this generative models alone. But if you think about it, now that we have these two foundation models together, they really can make things different, right. So we have predictor; we have the generator; you have a very good idea generator. And you have a very good goalkeeper. And you put them together. They form a loop. And now you can use this loop to design materials really quickly. So I would say to me, now, when I think about it, its really the combo that makes these set of tools different.KALTER: I know that Ive spoken with both of you recently about how theres so much excitement around this, and its clear that were on the precipice of thisas both of you have called ita paradigm shift. And Microsoft places a very strong emphasis on ensuring that its innovations are grounded in reality and capable of addressing real-world problems. So with that in mind, how do you balance the excitement of scientific exploration with the practical challenges of implementation? Tian, do you want to take this?XIE: Yeah, I think this is a very, very important point, because as there are so many hypes around AI that is happening right now, right. We must be very, very careful about the claims that we are making so that people will not have unrealistic expectations, right, over how these models can do. So for MatterGen, were pretty careful about that. Were trying to, basically, were trying to say that this is an early stage of generative AI in materials where this model will be improved over time quite significantly, but you should not say, oh, all the materials generated by MatterGen is going to be amazing. Thats not what is happening today. So we try to be very careful to understand how far MatterGen is already capable of designing materials with real-world impact. So therefore, we went all the way to synthesize one material that was generated by MatterGen. So this material we generated is called tantalum chromium oxide1. So this is a new material. It has not been discovered before. And it was generated by MatterGen by conditioning a bulk modulus equal to 200 gigapascal. Bulk modulus is, like, the compressiveness of the material. So we end up measuring the experimental synthesized material experimentally, and the measured bulk modulus is 169 gigapascal, which is within 20% of error. So this is a very good proof concept, in our point of view, to show that, oh, you can actually give it a prompt, right, and then MatterGen can generate a material, and the material actually have the property that is very close to your target. But its still a proof of concept. And were still working to see how MatterGen can design materials that are much more useful with a much broader range of applications. And Im sure that there will be more challenges we are seeing along the way. But were looking forward to further working with our experimental partners to, kind of, push this further. And also working with MatterSim, right, to see how these two tools can be used to design really useful materials and bringing this into real-world impact.LU: Yeah, Tian, I think thats very well said. Its not really only for MatterGen. For MatterSim, were also very careful, right. So we really want to make sure that people understand how these models really behave under their instructions and understand, like, what they can do and they cannot do. So I think one thing that we really care about is that in the next few, maybe one or two years, we want to really work with our experimental partners to make this realistic materials, like, in different areas so that we can, even us, can really better understand the limitations and at the same time explore the forefront of materials science to make this excitement become true.KALTER: Ziheng, could you give us a concrete example of what exactly MatterSim is capable of doing?LU: Now MatterSim can really do, like, whatever you have on a potential energy surface. So what that means is, like, anything that can be simulated with the energy and forces, stresses alone. So to give you an example, we can compute the first example would be the stability of a material. So basically, you input a structure, and from the energies of the relaxed structures, you can really tell whether the material is likely to be stable, like, the composition, right. So another example would be the thermal conductivity. Thermal conductivity is like a fundamental property of materials that tells you how fast heat can transfer in the material, right. So for MatterSim, it can really simulate how fast this heat can go through your diamond, your graphene, your copper, right. So basically, those are two examples. So these examples are based on energies and forces alone. But there are things MatterSim cannot doat least for now. For example, you cannot really do anything related to electronic structures. So you cannot really compute the light absorption of a semitransparent material. That would be a no-no for now.KALTER: Its clear from speaking with researchers, both from MatterSim and MatterGen, that despite these very rapid advancements in technology, you take very seriously the responsibility to consider the broader implications of the challenges that are still ahead. How do you think about the ethical considerations of creating entirely new materials and simulating their properties, particularly in terms of things like safety, sustainability, and societal impact?XIE: Yeah, thats a fantastic question. So its extremely important that we are making sure that these AI tools, they are not misused. A potential misuse, right, as you just mentioned, is that people begin to use these AI toolsMatterGen, MatterSimto, kind of, design harmful materials. There was actually extensive discussion over how generative AI tools that was originally purposed for drug design can be then misused to create bioweapons. So at Microsoft, we take this very seriously because we believe that when we create new technologies, you must also ensure that the technology is used responsibly. So we have an extensive process to ensure that all of our models respect those ethical considerations. In the meantime, as you mentioned, maybe sustainability and the societal impact, right, so theres a huge amount these AI toolsMatterGen, MatterSimcan do for sustainability because a lot of the sustainability challenges, they are really, at the end, materials design challenges, right. So therefore, I think that MatterGen and MatterSim can really help with that in solving, in helping us to alleviate climate change and having positive societal impact for the broader society.KALTER: And, Ziheng, how about from a simulation standpoint?LU: Yeah, I think Tian gave a very good, like, description. At Microsoft, we are really careful about these ethical, like, considerations. So I would add a little bit on the more, like, the bright side of things. Like, so for MatterSim, like, it really carries out these simulations at atomic scales. So one thing you can think about is really the educational purpose. So back in my bachelor and PhD period, so I would sit, like, at the table and really grab a pen to really deal with those very complex equations and get into those statistics using my pen. Its really painful. But now with MatterSim, these simulation tools at atomic level, what you can do is to really simulate the reactions, the movement of atoms, at atomic scale in real time. You can really see the chemical reactions and see the statistics. So you can get really the feeling, like very direct feeling, of how the system works instead of just working on those toy systems with your pen. I think its going to be a very good educational tool using MatterSim, yeah. Also MatterGen. MatterGen as, like, a generative tool and generating those i.i.d. (independent and identically distributed) distributions, it will be a perfect example to show the students how the Boltzmann distribution works. I think, Tian, you will agree with that, right?XIE: 100%. Yeah, I really, really like the example that Ziheng mentioned about the educational purposes. I still remember, like, when I was, kind of, learning material simulation class, right. So everything is DFT. You, kind of, need to wait for an hour, right, for getting some simulation. Maybe then youll make some animation. Now you can do this in real time. This is, like, a huge step forward, right, for our young researchers to, kind of, gaining a sense, right, about how atoms interact at an atomic level.LU: Yeah, and the results are really, I mean, true; not really those toy models. I think its going to be very exciting stuff.KALTER: And, Tian, Im directing this question to you, even though, Ziheng, Im sure you can chime in, as well. But, Tian, I know that you and I have previously discussed this specifically. I know that you said back in, you know, 2017, 2018, that you knew an AI-based approach to materials science was possible but that even you were surprised by how far the technology has come so fast in aiding this area. What is the status of these tools right now? Are they in use? And if so, who are they available to? And, you know, whats next for them?XIE: Yes, this is a fantastic question, right. So I think for AI generative tools like MatterGen, as I said many times earlier, its still in its early stages. MatterGen is the first tool that we managed to show that generative AI can enable very broad property-guided generation, and we have managed to have experimental validation to show its possible. But it will take more work to show, OK, it can actually design batteries, can design solar cells, right. It can design really useful materials in these broader domains. So this is, kind of, exactly why we are now taking a pretty open approach with MatterGen. We make our code, our training data, and model weights available to the general public. Were really hoping the community can really use our tools to the problem that they care about and even build on top of that. So in terms of what next, I always like to use what happened with generative AI for drugs, right, to kind of predict how generative AI will impact materials. Three years ago, there is a lot of research around generative model for drugs, first coming from the machine learning community, right. So then all the big drug companies begin to take notice, and then there are, kind of, researchers in these drug companies begin to use these tools in actual drug design processes. From my colleague, Marwin Segler, because he, kind of, works together with Novartis in Microsoft and Novartis collaboration, he has been basically telling me that at the beginning, all the chemists in the drug companies, theyre all very suspicious, right. The molecules generated by these generative models, they all look a bit weird, so they dont believe this will work. But once these chemists see one or two examples that actually turns out to be performing pretty well from the experimental result, then they begin to build more trust, right, into these generative AI models. And today, these generative AI tools, they are part of the standard drug discovery pipeline that is widely used in all the drug companies. That is today. So I think generative AI for materials is going through a very similar period. People will have doubts; people will have suspicions at the beginning. But I think in three years, right, so it will become a standard tool over how people are going to design new solar cells, design new batteries, and many other different applications.KALTER: Great. Ziheng, do you have anything to add to that?LU: So actually for MatterSim, we released the model, I think, back in last year, December. I mean, both the weights and the models, right. So were really grateful how much the community has contributed to the repo. And now, I mean, we really welcome the community to contribute more to both MatterSim and MatterGen via our open-source code bases. So, I mean, the community effort is really important, yeah.KALTER: Well, it has been fascinating to pick your brains, and as we close, you know, I know that youre both capable of quite a bit, which you have demonstrated. I know that asking you to predict the future is a big ask, so I wont explicitly ask that. But just as a fun thought exercise, lets fast-forward 20 years and look back. How have MatterGen and MatterSim and the big ideas behind them impacted the world, and how are people better off because of how you and your teams have worked to make them a reality? Tian, you want to start?XIE: Yeah, I think one of the biggest challenges our human society is going to face, right, in the next 20 years is going to be climate change, right, and there are so many materials design problems people need to solve in order to properly handle climate change, like finding new materials that can absorb CO2 from atmosphere to create a carbon capture industry or have a battery materials that is able to do large-scale energy grid storage so that we can fully utilizing all the wind powers and the solar power, etc., right. So if you want me to make one prediction, I really believe that these AI tools, like MatterGen and MatterSim, is going to play a central role in our humans ability to design these new materials for climate problems. So therefore in 20 years, I would like to see we have already solved climate change, right. We have large-scale energy storage systems that was designed by AI that is basically that we have removed all the fossil fuels, right, from our energy production, and for the rest of the carbon emissions that is very hard to remove, we will have a carbon capture industry with materials designed by AI that absorbs the CO2 from the atmosphere. Its hard to predict exactly what will happen, but I think AI will play a key role, right, into defining how our society will look like in 20 years.LU: Tian, very well said. So I think instead of really describing the future, I would really quote a science fiction scene in Iron Man. So basically in 20 years, I will say when we want to really get a new material, we will just sit in an office and say, Well, J.A.R.V.I.S., can you design us a new material that really fits my newest MK 7 suit? That will be the end. And it will run automatically, and we get this auto lab running, and all those MatterGen and MatterSim, these AI models, running, and then probably in a few hours, in a few days, we get the material.KALTER: Well, I think I speak for many people from several industries when I say that I cannot wait to see what is on the horizon for these projects. Tian and Ziheng, thank you so much for joining us on Ideas. Its been a pleasure.[MUSIC]XIE: Thank you so much.LU: Thank you.[MUSIC FADES]

Over the past year, our work on AutoGen has highlighted the transformative potential of agentic AI and multi-agent applications. Today, we are excited to announce AutoGen v0.4, a significant milestone informed by insights from our community of users and developers. This update represents a complete redesign of the AutoGen library, developed to improve code quality, robustness, generality, and scalability in agentic workflows.The initial release of AutoGen generated widespread interest in agentic technologies. At the same time, users struggled with architectural constraints, an inefficient API compounded by rapid growth, and limited debugging and intervention functionality. Feedback highlighted the need for stronger observability and control, more flexible multi-agent collaboration patterns, and reusable components. AutoGen v0.4 addresses these issues with its asynchronous, event-driven architecture.This update makes AutoGen more robust and extensible, enabling a broader range of agentic scenarios. The new framework includes the following features, inspired by feedback from both within and outside Microsoft.Asynchronous messaging: Agents communicate through asynchronous messages, supporting both event-driven and request/response interaction patterns.Modular and extensible: Users can easily customize systems with pluggable components, including custom agents, tools, memory, and models. They can also build proactive and long-running agents using event-driven patterns.Observability and debugging: Built-in metric tracking, message tracing, and debugging tools provide monitoring and control over agent interactions and workflows, with support for OpenTelemetry for industry-standard observability.Scalable and distributed: Users can design complex, distributed agent networks that operate seamlessly across organizational boundaries.Built-in and community extensions: The extensions module enhances the frameworks functionality with advanced model clients, agents, multi-agent teams, and tools for agentic workflows. Community support allows open-source developers to manage their own extensions.Cross-language support: This update enables interoperability between agents built in different programming languages, with current support for Python and .NET and additional languages in development.Full type support: Interfaces enforce type checks at build time, improving robustness and maintaining code quality.Spotlight: blog postGraphRAG auto-tuning provides rapid adaptation to new domainsGraphRAG uses LLM-generated knowledge graphs to substantially improve complex Q&A over retrieval-augmented generation (RAG). Discover automatic tuning of GraphRAG for new datasets, making it more accurate and relevant.Read moreOpens in a new tab New AutoGen frameworkAs shown in Figure 1, the AutoGen framework features a layered architecture with clearly defined responsibilities across the framework, developer tools, and applications. The framework comprises three layers: core, agent chat, and first-party extensions.Core: The foundational building blocks for an event-driven agentic system.AgentChat:A task-driven, high-level API built on the core layer, featuring group chat, code execution, pre-built agents, and more. This layer is most similar to AutoGen v0.2 (opens in new tab), making it the easiest API to migrate to.Extensions: Implementations of core interfaces and third-party integrations, such as the Azure code executor and OpenAI model client.Figure 1. The v0.4 update introduces a cohesive AutoGen ecosystem that includes the framework, developer tools, and applications. The frameworks layered architecture clearly defines each layers functionality. It supports both first-party and third-party applications and extensions.In addition to the framework, AutoGen 0.4 includes upgraded programming tools and applications, designed to support developers in building and experimenting with AutoGen.AutoGen Bench: Enables developers to benchmark their agents by measuring and comparing performance across tasks and environments.AutoGen Studio: Rebuilt on the v0.4 AgentChat API, this low-code interface enables rapid prototyping of AI agents. It introduces several new capabilities:Real-time agent updates: View agent action streams in real time with asynchronous, event-driven messages.Mid-execution control: Pause conversations, redirect agent actions, and adjust team composition. Then seamlessly resume tasks.Interactive feedback through the UI: Add a UserProxyAgent to enable user input and guidance during team runs in real time.Message flow visualization: Understand agent communication through an intuitive visual interface that maps message paths and dependencies.Drag-and-drop team builder: Design agent teams visually using an interface for dragging components into place and configuring their relationships and properties.Third-party component galleries: Import and use custom agents, tools, and workflows from external galleries to extend functionality.Magentic-One: A new generalist multi-agent application to solve open-ended web and file-based tasks across various domains. This tool marks a significant step toward creating agents capable of completing tasks commonly encountered in both work and personal contexts.Migrating to AutoGen v0.4We implemented several measures to facilitate a smooth upgrade from the previous v0.2 API, addressing core differences in the underlying architecture.First, the AgentChat API maintains the same level of abstraction as v0.2, making it easy to migrate existing code to v0.4. For example, AgentChat offers an AssistantAgent and UserProxy agent with similar behaviors to those in v0.2. It also provides a team interface with implementations like RoundRobinGroupChat and SelectorGroupChat, which cover all the capabilities of the GroupChat class in v0.2. Additionally, v0.4 introduces many new functionalities, such as streaming messages, improved observability, saving and restoring task progress, and resuming paused actions where they left off. For detailed guidance, refer to the migration guide (opens in new tab).Looking forwardThis new release sets the stage for a robust ecosystem and strong foundation to drive advances in agentic AI application and research. Our roadmap includes releasing .NET support, introducing built-in, well-designed applications and extensions for challenging domains, and fostering a community-driven ecosystem. We remain committed to the responsible development of AutoGen and its evolving capabilities.We encourage you to engage with us on AutoGens Discord server (opens in new tab) and share feedback on the official AutoGen repository (opens in new tab) via GitHub Issues. Stay up to date with frequent AutoGen updates via X.AcknowledgmentsWe would like to thank the many individuals whose ideas and insights helped formalize the concepts introduced in this release, including Rajan Chari, Ece Kamar, John Langford, Ching-An Chen, Bob West, Paul Minero, Safoora Yousefi, Will Epperson, Grace Proebsting, Enhao Zhang, and Andrew Ng.Opens in a new tab

Welcome to Research Focus, a series of blog posts that highlights notable publications, events, code/datasets, new hires and other milestones from across the research community at Microsoft.NEW RESEARCHThe Compute Express Link (CXL) open standard interconnect enables integration of diverse types of memory into servers via its byte-addressable SerDes links. To fully utilize CXL-based heterogeneous memory systems (which combine different types of memory with varying access speeds), its necessary to implement efficient memory tieringa strategy to manage data placement across memory tiers for optimal performance. Efficiently managing these memory systems is crucial, but has been challenging due to the lack of precise and efficient tools for understanding how memory is accessed.In a recent paper: NeoMem: Hardware/Software Co-Design for CXL-Native Memory Tiering researchers from Microsoft propose a novel solution which features a hardware/software co-design to address this problem. NeoMem offloads memory profiling functions to CXL device-side controllers, integrating a dedicated hardware unit called NeoProf, which monitors memory accesses and provides the operating system (OS) with crucial page hotness statistics and other system state information. On the OS kernel side, the researchers designed a revamped memory-tiering strategy, enabling accurate and timely hot page promotion based on NeoProf statistics. Implemented on a real FPGA-based CXL memory platform and Linux kernel v6.3, NeoMem demonstrated 32% to 67% geomean speedup over several existing memory tiering solutions.Read the paperNEW RESEARCHPlanning and conducting chemical syntheses is a significant challenge in the discovery of functional small molecules, which limits the potential of generative AI for molecular inverse design. Although early machine learning-based retrosynthesis models have shown the ability to predict reasonable routes, they are less accurate for infrequent, yet important reactions.In a recent paper: Chimera: Accurate retrosynthesis prediction by ensembling models with diverse inductive biases, researchers from Microsoft and external colleagues address this limitation, with a new framework for building highly accurate reaction models. Chimera incorporates two newly developed models, each achieving state-of-the-art performance in their respective categories. Evaluations by PhD-level organic chemists show that Chimeras predictions are preferred for their higher quality compared to baseline models.The researchers further validate Chimeras robustness by applying its largest-scale model to an internal dataset from a major pharmaceutical company, demonstrating its ability to generalize effectively under distribution shifts. This new framework shows the potential to substantially accelerate the development of even more accurate and versatile reaction prediction models.Read the paperMicrosoft research podcastAbstracts: August 15, 2024Advanced AI may make it easier for bad actors to deceive others online. A multidisciplinary research team is exploring one solution: a credential that allows people to show theyre not bots without sharing identifying information. Shrey Jain and Zo Hitzig explain.Listen nowOpens in a new tab NEW RESEARCHIn bioinformatics and computational biology, data analysis often involves chaining command-line programs developed by specialized teams at different institutions. These tools, which vary widely in age, software stacks, and dependencies, lack a common programming interface, which makes integration, workflow management and reproducibility challenging.A recent article (opens in new tab) emphasizes the development, adoption and implementation of the Global Alliance for Genomics and Health (GA4GH) Task Execution Service (TES) API, created in collaboration with researchers at Microsoft and other institutions. The TES API offers a unified schema and interface for submitting and managing tasks, seamlessly bridging gaps between on-premises high-performance and high-throughput computing systems, cloud platforms, and hybrid infrastructures. Its flexibility and extensibility have already made it a critical asset for applications ranging from federated data analysis to load balancing across multi-cloud systems.Adopted by numerous service providers and integrated into several workflow engines, TES empowers researchers to execute complex computational tasks through a single, abstracted interface. This eliminates compatibility hurdles, accelerates research timelines, reduces costs and enables compute to data solutionsessential for tackling the challenges of distributed data analysis.Read the paperNEW RESEARCHIncreasing use of code agents for AI-assisted coding and software development has brought safety and security concerns, such as generating or executing malicious code, which have become significant barriers to real-world deployment of these agents.In a recent paper: RedCode: Risky Code Execution and Generation Benchmark for Code Agents, published at NeurIPS 2024, researchers from Microsoft and external colleagues propose comprehensive and practical evaluations on the safety of code agents. RedCode is an evaluation platform with benchmarks grounded in four key principles: real interaction with systems, holistic evaluation of unsafe code generation and execution, diverse input formats, and high-quality safety scenarios and tests.This research evaluated three agents based on various large language models (LLMs), providing insights into code agents vulnerabilities. For instance, results showed that agents are more likely to reject executing unsafe operations on the operating system. Unsafe operations described in natural text lead to a lower rejection rate than those in code format. Additional evaluations revealed that more capable base models and agents with stronger overall coding abilities, such as GPT-4, tend to produce more sophisticated harmful software.These findings highlight the need for stringent safety evaluations for diverse code agents. The underlying dataset and related code are publicly available at https://github.com/AI-secure/RedCode (opens in new tab).Read the paperNEW RESEARCHAlthough large language models (LLMs) excel at language-focused tasks like news writing, document summarization, customer service, and supporting virtual assistants, they can face challenges when it comes tolearning and inference on numeric and structured industry data, such as tabular and time series data. To address these issues, researchers from Microsoft propose a new approach to building industrial foundation models (IFMs). As outlined in a recent blog post, they have successfully demonstrated the feasibility of cross-domain universal in-context learning on tabular data and the significant potential it could achieve.The researchers designed Generative Tabular Learning (opens in new tab)(GTL), a new framework that integrates multi-industry zero-shot and few-shot learning capabilities into LLMs. This approach allows the models to adapt and generalize to new fields, new data, and new tasks more effectively, flexibly responding to diverse data science tasks. This technical paradigm has been open-sourced (opens in new tab)to promote broader use.Read the paperMicrosoft Research in the newsMicrosofts smaller AI model beats the big guys: Meet Phi-4, the efficiency kingDecember 12, 2024Microsoft launched a new artificial intelligence model today that achieves remarkable mathematical reasoning capabilities while using far fewer computational resources than its larger competitors. Microsoft researcher Ece Kamar discusses the future of AI agents in 2025Tech Brew | December 12, 2024With AI agents widely expected to take off in 2025, the director of Microsofts AI Frontiers lab weighs in on the future of this technology, the safeguards needed, and the year ahead in AI research. A new frontier awaits computing with lightDecember 12, 2024In the guts of a new type of computer, a bunch of tiny LEDs emit a green glow. Those lights have a job to do. Theyre performing calculations. Right now, this math is telling the computer how to identify handwritten images of numbers. The computer is part of a research program at Microsoft. View more news and awards Opens in a new tab

Transcript[TEASER][MUSIC PLAYS UNDER DIALOGUE]MADELEINE DAEPP: Last summer, I was working on all of these like pro-democracy applications, trying to build out, like, a social data collection tool with AI, all this kind of stuff. And I went to the elections workshop that the Democracy Forward team at Microsoft had put on, and Dave Leichtman, who, you know, was the MC of that work, was really talking about how big of a global elections year 2024 was going to be. Over 70 countries around the world. And, you know, were coming from Microsoft Research, where we were so excited about this technology. And then, all of a sudden, I was at the elections workshop, and I thought, oh no, [LAUGHS] like, this is not good timing.ROBERT OSAZUWA NESS: What are we really talking about in the context of deepfakes in the political context, elections context? Its deception, right. Im trying to use this technology to, say, create some kind of false record of events in order to convince people that something happened that actually did not happen. And so that goal of deceiving, of creating a false record, thats kind of how I have been thinking about deepfakes in contrast to the broader category of generative AI.[TEASER ENDS]GINNY BADANES: Welcome to Ideas, a Microsoft Research Podcast that dives deep into the world of technology research and the profound questions behind the code. In this series, well explore the technologies that are shaping our future and the big ideas that propel them forward.[MUSIC FADES]Im your guest host, Ginny Badanes, and I lead Microsofts Democracy Forward program, where weve spent the past year deeply engaged in supporting democratic elections around the world, including the recent US elections. We have been working on everything from raising awareness of nation-state propaganda efforts to helping campaigns and election officials prepare for deepfakes to protecting political campaigns from cyberattacks. Today, Im joined by two researchers who have also been diving deep into the impact of generative AI on democracy.Microsoft senior researchers Madeleine Daepp and Robert Osazuwa Ness are studying generative AIs influence in the political sphere with the goal of making AI systems more robust against misuse while supporting the development of AI tools that can strengthen democratic processes and systems. They spent time in Taiwan and India earlier this year, where both had big democratic elections. Madeleine and Robert, welcome to the podcast!MADELEINE DAEPP: Thanks for having us.ROBERT OSAZUWA NESS: Thanks for having us.BADANES: So I have so many questions for you allfrom how you conducted your research to what youve learnedand Im really interested in what you think comes next. But first, lets talk about how you got involved in this in the first place. Could you both start by telling me a little bit about your backgrounds and just what got you into AI research in the first place?DAEPP: Sure. So Im a senior researcher here at Microsoft Research in the Special Projects team. But I did my PhD at MIT in urban studies and planning. And I think a lot of folks hear that field and think, oh, you know, housing, like upzoning housing and figuring out transportation systems. But it really is a field thats about little d democracy, right. About how people make choices about shared public spaces every single day. You know, I joined Microsoft first off to run this, sort of, technology deployment in the city of Chicago, running a low-cost air-quality-sensor network for the city. And when GPT-4 came out, you know, first ChatGPT, and then we, sort of, had this big recognition of, sort of, how well this technology could do in summarizing and in representing opinions and in making sense of big unstructured datasets, right. I got actually very excited. Like, I thought this could be used for town planning processes. [LAUGHS] Like, I thought we could I had a whole project with a wonderful intern, Eva Maxfield Brown, looking at, can we summarize planning documents using AI? Can we build out policies from conversations that people have in shared public spaces? And so that was very much the impetus for thinking about how to apply and build things with this amazing new technology in these spaces.BADANES: Robert, I think your background is a little bit different, yet you guys ended up in a similar place. So how did you get there?NESS: Yeah, so Im also on Special Projects, Microsoft Research. My work is focusing on large language models, LLMs. And, you know, so I focus on making these models more reliable and controllable in real-world applications. And my PhD is in statistics. And so I focus a lot on using just basic bread-and-butter statistical methods totry and control and understand LLM behavior. So currently, for example, Im leading a team of engineers and running experiments designed to find ways to enhance a graphical approach to combining information retrieval in large language models. I work on statistical tests for testing significance of adversarial attacks on these models.BADANES: Wow.NESS: So, for example, if you find a way to trick one of these models into doing something its not supposed to do, I make sure that its not, like, a random fluke; that its something thats reproducible. And I also work at this intersection between generative AI and, you know, Bayesian stuff, causal inference stuff. And so I came at looking at this democracy work through an alignment lens. So alignment is this task in AI of making sure these models align with human values and goals. And what I was seeing was a lot of research in the alignment space was viewing it as a technical problem. And, you know, as a statistician, were trained to consult, right. Like, to go to the actual stakeholders and say, hey, what are your goals? What are your values? And so this democracy work was an opportunity to do that in Microsoft Research and connected with Madeleine. So she was planning to go to Taiwan, and kind of from a past life, I wanted to become a trade economist and learned Mandarin. And so I speak fluent Mandarin and seemed like a good matchup of our skill sets BADANES: Yeah.NESS: and interests. And so thats, kind of, how we got started.BADANES: So, Madeleine, you brought the two of you together, but what started it for you? This podcast is all about big ideas. What sparked the big idea to bring this work that youve been doing on generative AI into the space of democracy and then to go out and find Robert and match up together?DAEPP: Yeah, well, Ginny, it was you. [LAUGHS] It was actually your team.BADANES: I didnt plant that! [LAUGHS]DAEPP: So, you know, I think last summer, I was working on all of these like pro-democracy applications, trying to build out, like, a social data collection tool with AI, all this kind of stuff. And I went to the elections workshop that the Democracy Forward team at Microsoft had put on, and Dave Leichtman, who, you know, was the MC of that work, was really talking about how big of a global elections year 2024 was going to be, that thishe was calling it Votorama. You know, that term didnt take off. [LAUGHTER] The term that has taken off is biggest election year in history, right. Over 70 countries around the world. And, you know, were coming from Microsoft Research, where we were so excited about this technology. Like, when it started to pass theory of mind tests, right, which is like the ability to think about how other people are thinking, like, we were all like, oh, this is amazing; this opens up so many cool application spaces, right. When it was, like, passing benchmarks for multilingual communication, again, like, we were so excited about the prospect of building out multilingual systems. And then, all of a sudden, I was at the elections workshop, and I thought, oh no, [LAUGHS] this is not good timing.BADANES: Yeah DAEPP: And because so much of my work focuses on, you know, building out computer science systems like, um, data science systems or AI systems but with communities in the loop, I really wanted to go to the folks most affected by this problem. And so I proposed a project to go to Taiwan and to study one of the it was the second election of 2024. And Taiwan is known to be subject to more external disinformation than any other place in the world. So if you were going to see something anywhere, you would see it there. Also, it has amazing civil society response so really interesting people to talk to. But I do not speak, Chinese, right. Like, I dont have the context; I dont speak the language. And so part of my process is to hire a half-local team. We had an amazing interpreter, Vickie Wang, and then a wonderful graduate student, Ti-Chung Cheng, who supported this work. But then also my team, Special Projects, happened to have this person who, like, not only is a leading AI researcher publishing in NeurIPS, like building out these systems, but who also spoke Chinese, had worked in technology security, and had a real understanding of international studies and economics as well as AI. And so for me, like, finding Robert as a collaborator was kind of a unicorn moment.BADANES: So it sounds like it was a match made in heaven of skill sets and abilities. Before we get into what you all found there, which I do want to get into, I first think its helpfulI dont know, when were dealing with these, like, complicated issues, particularly things that are moving and changing really quickly, sometimes I found its helpful to agree on definitions and sort of say, this is what we mean when we say this word. And that helps lead to understanding. So while I know that this research is about more than deepfakesand well talk about some of the things that are more than deepfakesI am curious how you all define that term and how you think of it. Because this is something that I think is constantly moving and changing. So how have you all been thinking about the definition of that term?NESS: So Ive been thinking about it in terms of the intention behind it, right. We say deepfake, and I think colloquially that means kind of all of generative AI. Thats a bit unfortunate because there are things that are you know, you can use generative AI to generate cartoons BADANES: Right.NESS: or illustrations for a childrens book. And so in thinking about what are we really talking about in the context of deepfakes in the political context, elections context, its deception, right. Im trying to use this technology to, say, create some kind of false record of events, say, for example, something that a politician says, in order to convince people that something happened that actually did not happen.BADANES: Right.NESS: And so that goal of deceiving, of creating a false record, thats kind of how I have been thinking about deepfakes in contrast to the broader category of generative AI and deepfakes in terms of being a malicious use case. There are other malicious use cases that dont necessarily have to be deceptive, as well, as well as positive use cases.BADANES: Well, that really, I mean, that resonates with me because what we found was when you use the term deceptionor another term we hear a lot that I think works is fraudthat resonates with other people, too. Like, that helps them distinguish between neutral uses or even positive uses of AI in this space and the malicious use cases, though to your point, I suppose theres probably even deeper definitions of what malicious use could look like. Are you finding that distinction showing up in your work between fraud and deception in these use cases? Is that something that has been coming through?DAEPP: You know, we didnt really think about the term fraud until we started prepping for this interview with you.financially was not OK, right. Thats fraud. Using AI for the purposes of nudifying, like removing somebodys clothes and then sextorting them, right, extorting them for money out of fear that this would be shared, like, that was not OK. And those are such clear lines. And it was clear that theres a set of uses of generative AI also in the political space, you know, of saying this person said something that they didnt, BADANES: Mm-hmm.DAEPP: of voter suppression, that in general, theres a very clear line that when it gets into that fraudulent place, when it gets into that simultaneously deceptive and malicious space, thats very clearly a no-go zone.NESS: Oftentimes during this research, I found myself thinking about this dichotomy in cybersecurity of state actors, or broadly speaking, kind of, political actors, versus criminals.BADANES: Right.NESS: And its important to understand the distinction because criminals are typically trying to target targets of opportunity and make money, while state-sponsored agents are willing to spend a lot more money and have very specific targets and have a very specific definition of success. And so, like, this fraud versus deception kind of feels like that a little bit in the sense that fraud is typically associated with criminal behavior, while, say, I might put out deceptive political messaging, but it might fall within the bounds of free speech within my country.BADANES: Right, yeah.NESS: And so this is not to say I disagree with that, but it just, actually, that it could be a useful contrast in terms of thinking about the criminal versus the political uses, both legitimate and illegitimate.BADANES: Well, I also think those of us who work in the AI space are dealing in very complicated issues that the majority of the world is still trying to understand. And so any time you can find a word that people understand immediately in order to do the, sort of, storytelling: the reason that we are worried about deepfakes in elections is because we do not want voters to be defrauded. And that, we find really breaks through because people understand that term already. Thats a thing that they already know that they dont want to be; they do not want to be defrauded in their personal life or in how they vote. And so that really, I found, breaks through. But as much as I have talked about deepfakes, I know that youand I know theres a lot of interest in talking about deepfakes when we talk about this subjectbut I know your research goes beyond that. So what other forms of generative AI did you include in your research or did you encounter in the effort that you were doing both in Taiwan and India?DAEPP: Yeah. So let me tell you just, kind of, a big overview of, like, our taxonomy. Because as you said, like, so much of this is just about finding a word, right. Like, so much of it is about building a shared vocabulary so that we can start to have these conversations. And so when we looked at the political space, right, elections, so much of what it means to win an election is kind of two things. Its building an image of a candidate, right, or changing the image of your opposition and telling a story, right.BADANES: Mm-hmm.DAEPP: And so if you think about image creation, of course, there are deepfakes. Like, of course, there are malicious representations of a person. But we also saw a lot of what were calling auth fakes, like authorized fakes, right. Candidates who would actually go to a consultancy and, like, get their bodies scanned so that videos could be made of them. Theyd get their voices, a bunch of snippets of their voices, recorded so that then there could be personalized phone calls, right. So these are authorized uses of their image and likeness. Then we saw a term Ive heard in, sort of, the ether is soft fakes. So again, likenesses of a candidate, this time not necessarily authorized but promotional. They werent people on TwitterI guess, Xon Instagram, they were sharing images of the candidate that they supported that were really flattering or silly or, you know, just really sort of in support of that person. So not with malicious intent, right, with promotional intent. And then the last one, and this, I think, was Roberts term, but in this image creation category, you know, one thing we talked about was just the way that people were also making fun of candidates. And in this case, this is a bit malicious, right. Like, theyre making fun of people; theyre satirizing them. But its not deceptive because, BADANES: Right DAEPP: you know, often it has that hyper-saturated meme aesthetic. Its very clearly AI or just, you know, per like, sort of, US standards for satire, like, a reasonable person would know that it was silly. And so Robert said, you know, oh, these influencers, theyre not trying to deceive people; like, theyre not trying to lie about candidates. Theyre trying to roast them. [LAUGHTER] And so we called it a deep roast. So thats, kind of, the images of candidates. I will say we also looked at narrative building, and there, one really important set of things that we saw was what we call text to b-roll. So, you know, a lot of folks think that you cant really make AI videos because, like, Sora isnt out yet[1]. But in fact, what there is a lot of is tooling to, sort of, use AI to pull from stock imagery and b-roll footage and put together a 90-second video. You know, it doesnt look like AI; its a real video. So text to b- roll, AI pasta? So if you know the threat intelligence space, theres this thing called copy pasta, where people just BADANES: Sure.DAEPP: its just a fun word for copy-paste. People just copy-paste terms in order to get a hashtag trending. And we talked to an ex-influencer who said, you know, were using AI to do this. And I asked him why. And he said, well, you know, if you just do copy-paste, the fact-checkers catch it. But if you use AI, they dont. And so AI pasta. And theres also some research showing that this is potentially more persuasive than copy-paste BADANES: Interesting.DAEPP: because people think theres a social consensus. And then the last one, this is my last of the big taxonomy, and, Robert, of course, jump in on anything you want to go deeper on, but Fake News 2.0. You know, Im sure youve seen this, as well. Just this, like, creation of news websites, like entire new newspapers that nobodys ever heard of. AI avatars that are newscasters. And this is something that was happening before. Like, theres a long tradition of pretending to be a real news pamphlet or pretending to be a real outlet. But theres some interesting work out of Patrick Warren at Clemson has looked at some of these and shown the quality and quantity of articles on these things has gotten a lot better and, you know, improves as a step function of, sort of, when new models come out.NESS: And then on the flip side, you have people using the same technologies but stated clearly that its AI generated, right. So we mentioned the AI avatars. In India, theres this theres Bhoomi, which is a AI news anchor for agricultural news, and it states there in clear terms that shes not real. But of course, somebody who wanted to be deceptive could use the same technology to portray something that looks like a real news broadcast that isnt. You know, and, kind of, going back, Madeleine mentioned deep roasts, right, so, kind of, using this technology to create satirical depictions of, say, a political opponent. Somebody, a colleague, sent something across my desk. It was a Douyin accountso Douyin is the version of TikTok thats used inside China; BADANES: OK.NESS: same company, but its the internal version of TikTokthat was posting AI-generated videos of politicians in Taiwan. And these were excellent, real good-quality AI-generated deepfakes of these politicians. But some of them were, first off, on the bottom of all of them, it said, this is AI-generated content.BADANES: Oh.NESS: And some of them were, kind of, obviously meant to be funny and were clearly fake, like still images that were animated to make somebody singing a funny song, for example. A very serious politician singing a very silly song. And its a still image. Its not even, its not even BADANES: a video.NESS: like video.BADANES: Right, right.NESS: And so I messaged Puma Shen, who is one of the legislators in Taiwan who was targeted by these attacks, and I said, what do you think about this? And, you know, he said, yeah, they got me. [LAUGHTER] And I said, you know, do you think people believe this? I mean, there are people who are trying to debunk it. And he said, no, our supporters dont believe it, but, you know, people who support the other side or people who are apolitical, they might believe it, or even if it says its fakethey know its fakebut they might still say that, yeah, but this is something they would do, right. This is BADANES: Yeah, it fits the narrative. Yeah.NESS: it fits the narrative, right. And that, kind of, that really, you know, I had thought of this myself, but just hearing somebody, you know, whos, you know, a politician whos targeted by these attacks just saying that its, like, even if they believe its even if they know its fake, they still believe it because its something that they would do.BADANES: Sure.NESS: Thats, you know, as a form of propaganda, even relative to the canonical idea of deepfake that we have, this could be more effective, right. Like, just say its AI and then use it to, kind of, paint the picture of the opponent in any way you like.BADANES: Sure, and this gets into that, sort of, challenging space I think we find ourselves in right now, which is people dont know necessarily how to tell whats real or not. And the case youre describing, it has labeling, so that should tell you. But a lot of the content we come across online does not have labeling. And you cannot tell just based on your eyes whether images were generated by AI or whether theyre real. One of the things that I get asked a lot is, why cant we just build good AI to detect bad AI, right? Why dont we have a solution where I just take a picture and I throw it into a machine and it tells me thumbs-up or thumbs-down if this is AI generated or not? And the question around detection is a really tricky one. Im curious what you all think about, sort of, the question of, can detection solve this problem or not?NESS: So Ill mention one thing. So Madeleine mentioned an application of this technology called text to b-roll. And so what this is, technically speaking, what this is doing is youre taking real footage, you stick it in a database, its quote, unquote vectorized into these representations that the AI can understand, and then you say, hey, generate a video that illustrates this narrative for me. And you provide it the text narrative, and then it goes and pulls out a whole bunch of real video from a database and curates them into a short video that you could put on TikTok, for example. So this was a fully AI-generated product, but none of the actual content is synthetic.BADANES: Ah, right.NESS: So in that case, your quote, unquote AI detection tool is not going to work.DAEPP: Yeah, I mean, something that I find really fascinating any time that youre dealing with a sociotechnical system, righta technical system embedded in social contextis folks, you know, think that things are easy that are hard and things are hard that are easy, right. And so with a lot of the detections work, right, like if you put a deepfake detector out, you make that available to anyone, then what they can do is they can run a bunch of stuff by it, BADANES: Yeah.DAEPP: add a little bit of random noise, and then the deepfake detector doesnt work anymore. And so that detection, actually, technically becomes an arms race, you know. And were seeing now some detectors that, like, you know, work when youre not looking at a specific image or a specific piece of text but youre looking at a lot all at once. That seems more promising. But, just, this is a very, very technically difficult problem, and that puts us as researchers in a really tricky place because, you know, youre talking to folks who say, why cant you just solve this? If you put this out, then you have to put the detector out. And were like, thats actually not, thats not a technically feasible long-term solution in this space. And the solutions are going to be social and regulatory and, you know, changes in norms as well as technical solutions that maybe are about everything outside of AI, right.BADANES: Yeah.DAEPP: Not about fixing the AI system but fixing the context within which its used.BADANES: Its not just a technological solution. Theres more to it. Robert?NESS: So if somebody were to push back there, they could say, well, great; in the long term, maybe its an arms race, but in the short term, right, we can have solutions out there that, you know, at least in the next election cycle, we could maybe prevent some of these things from happening. And, again, kind of harkening back to cybersecurity, maybe if you make it hard enough, only the really dedicated, really high-funded people are going to be doing it rather than, you know, everybody who wants to throw a bunch of deepfakes on the internet. But the problem still there is that it focuses really on video and images, right.BADANES: Yeah. What about audio?NESS: What about audio? And what about text? So BADANES: Yeah. Those are hard. I feel like weve talked a lot about definitions and theoretical, but I want to make sure we talk more about what you guys saw and researched and understood on the ground, in particular, your trips to India and Taiwan and even if you want to reflect on how those compare to the US environment. What did you actually uncover? What surprised you? What was different between those countries?DAEPP: Yeah, I mean, right, so Taiwan both of these places are young democracies. And thats really interesting, right. So like in Taiwan, for example, when people vote, they vote on paper. And anybody can go watch. Thats part of their, like, security strategies. Like, anyone around the world can just come and watch. People come from far. They fly in from Canada and Japan and elsewhere just to watch Taiwanese people vote. And then similarly in India, theres this rule where you have to be walking distance from your polling place, and so the election takes two months. And, like, your polling places move from place to place, and sometimes, it arrives on an elephant. And so these were really interesting places to, like, I as an American, just, like, found it very, very fascinating to and important to be outside of the American context. You know, we just take for granted that how we do democracy is how other people do it. But Taiwan was very much a joint, like, civil societygovernment everyday response to this challenge of having a lot of efforts to manipulate public opinion happening with, you know, real-world speeches, with AI, with anything that you can imagine. You know, and I think the Microsoft Threat Analysis Center released a report documenting some of the, sort of, video stuff[2]. Theres a use of AI to create videos the night before the election, things like this. But then India is really thinking of so India, right, its the worlds biggest democracy, right. Like, nearly a billion people were eligible to vote.BADANES: Yeah.NESS: And arguably the most diverse, right?DAEPP: Yeah, arguably the most diverse in terms of languages, contexts. And its also positioning itself as the AI laboratory for the Global South. And so folks, including folks at the MSR (Microsoft Research) Bangalore lab, are leaders in thinking about representing low-resource languages, right, thinking about cultural representation in AI models. And so there you have all of these technologists who are really trying to innovate and really trying to think about whats the next clever application, whats the next clever use. And so that, sort of, that taxonomy that we talked about, like, I think just every week, every interview, we, sort of, had new things to add because folks there were just constantly trying all different kinds of ways of engaging with the public.NESS: Yeah, I think for me, in India in particular, you know, India is an engineering culture, right. In terms of, like, the professional culture there, theyre very, kind of, engineering skewed. And so I think one of the bigger surprises for me was seeing people who were very experienced and effective campaign operatives, right, people who would go and, you know, hit the pavement; do door knocking; kind of, segment neighborhoods by demographics and voter block, these people were also, you know, graduated in engineering from an IIT (Indian Institute of Technology), BADANES: Sure.NESS: right, and so [LAUGHS] so they were happy to pick up these tools and leverage them to support their expertise in this work, and so some of the, you know, I think a lot of the narrative that we tell ourselves in AI is how its going to be, kind of, replacing people in doing their work. But what I saw in India was that people who were very effective had a lot of domain expertise that you couldnt really automate away and they were the ones who are the early adopters of these tools and were applying it in ways that I think were behind on in terms of, you know, ideas in the US.BADANES: Yeah, I mean, theres, sort of, this sentiment that AI only augments existing problems and can enhance existing solutions, right. So were not great at translation tools, but AI will make us much better at that. But that also can then be weaponized and used as a tool to deceive people, which propaganda is not new, right? Were only scaling or making existing problems harder, or adversaries are trying to weaponize AI to build on things theyve already been doing, whether thats cyberattacks or influence operations. And while the three of us are in different roles, we do work for the same company. And its a large technology company that is helping bring AI to the world. At the same time, I think there are some responsibilities when we look at, you know, bad actors who are looking to manipulate our products to create and spread this kind of deceptive media, whether its in elections or in other cases like financial fraud or other ways that we see this being leveraged. Im curious what you all heard from others when youve been doing your research and also what you think our responsibilities are as a big tech company when it comes to keeping actors from using our products in those ways.DAEPP: You know, when I started using GPT-4, one of the things I did was I called my parents, and I said, if you hear me on a phone call, BADANES: Yeah.DAEPP: like, please double check. Ask me things that only I would know. And when I walk around Building 99, which is, kind of, a storied building in which a lot of Microsoft researchers work, everybody did that call. We all called our parents.BADANES: Interesting.DAEPP: Or, you know, we all checked in. So just as, like, we have a responsibility to the folks that we care about, I think as a company, that same, sort of, like, raising literacy around the types of fraud to expect and how to protect yourself from themI think that gets back to that fraud space that we talked aboutand, you know, supporting law enforcement, sharing what needs to be shared, I think that without question is a space that we need to work in. I will say a lot of the folks we talked with, they were using Llama on a local GPU, right.BADANES: OK.DAEPP: They were using open-source models. They were sometimes they were testing out Phi. They would use Phi, Grok, Llama, like anything like that. And so that raises an interesting question about our guardrails and our safety practices. And I think there, we have an, like, our obligation and our opportunity actually is to set the standard, right. To say, OK, like, you know, if you use local Llama and it spouts a bunch of stuff about voter suppression, like, you can get in trouble for that. And so what does it mean to have a safe AI that wins in the marketplace, right? Thats an AI that people can feel confident and comfortable about using and one thats societally safe but also personally safe. And I think thats both a challenge and a real opportunity for us.BADANES: Yeah oh, go ahead, Robert, yeah NESS: Going back to the point about fraud. It was this year, in January, when that British engineering firm Arup, when somebody used a deepfake to defraud that company of about $25 million, BADANES: Yeah.NESS: their Hong Kong office. And after that happened, some business managers in Microsoft reached out to me regarding a major client who wanted to start red teaming. And by red teaming, I mean intentionally targeting your executives and employees with these types of attacks in order to figure out where your vulnerabilities as an organization are. And I think, yeah, it got me thinking like, wow, I would, you know, can we do this for my dad? [LAUGHS] Because I think that was actually a theme that came out from a lot of this work, which was, like, how can we empower the people who are really on the frontlines of defending democracy in some of these places in terms of the tooling there? So we talked about, say, AI detection tools, but the people who are actually doing fact-checking, theyre looking more than at just the video or the images; theyre actually looking at a, kind of, holistic taking a holistic view of the news story and doing some proper investigative journalism to see if something is fake or not.BADANES: Yeah.NESS: And so I think as a company who creates products, can we take a more of a product mindset to building tools that support that entire workflow in terms of fact-checking or investigative journalism in the context of democratic outcomes BADANES: Yeah.NESS: where maybe looking at individual deepfake content is just a piece of that.BADANES: Yeah, you know, I think theres a lot of parallels here to cybersecurity. Thats also what weve found, is this idea that, first of all, the no silver bullet, as we were talking about earlier with the detection piece. Like, you cant expect your system to be secure just because you have a firewall, right. You have to have this, like, defense in-depth approach where you have lots of different layers. And one of those layers has been on the literacy side, right. Training and teaching people not to click on a phishing link, understanding that they should scroll over the URL. Like, these are efforts that have been taken up, sort of, in a broad societal sense. Employers do it. Big tech companies do it. Governments do it through PSAs and other things. So theres been a concerted effort to get a population who might not have been aware of the fact that they were about to be scammed to now know not to click on that link. I think, you know, you raised the point about literacy. And I think theres something to be said about media literacy in this space. Its both AI literacyunderstanding what it isbut also understanding that people may try to defraud you. And whether that is in the political sense or in the financial sense, once you have that, sort of, skill set in place, youre going to be protected. One thing that Ive heard, though, as I have conversations about this challenge Ive heard a couple things back from people specifically in civil society. One is not to put the impetus too much on the end consumer, which I think Im hearing that we also recognize theres things that we as technology companies should be focusing on. But the other thing is the concern that in, sort of, the long run, were going to all lose trust in everything we see anyway. And Ive heard some people refer to that as the trust deficit. Have you all seen anything promising in the space to give you a sense around, can we ever trust what were looking at again, or are we actually just training everyone to not believe anything they see? Which I hope is not the case. I am an optimist. But Id love to hear what you all came across. Are there signs of hope here where we might actually have a place where we can trust what we see again?DAEPP: Yeah. So two things. There is this phenomenon called the liars dividend, right, BADANES: Sure, yeah.DAEPP: which is where that if you educate folks about how AI can be used to create fake clips, fake audio clips, fake videos, then if somebody has a real audio clip, a real video, they can claim that its AI. And I think we talk, you know, again, this is, like, in a US-centric space, we talk about this with politicians, but the space in which this is really concerning, I think, is war crimes, right BADANES: Oh, yeah.DAEPP: I think are these real human rights infractions where you can prevent evidence from getting out or being taken seriously. And we do see that right after invasions, for example, these days. But this is actually a space like, I just told you, like, oh, like, detection is so hard and not technically, like, thatll be an arms race! But actually, there is this wonderful project, Project Providence, that is a Microsoft collaboration with a company called Truepic that its, like, an app, right. And what happens is when you take a photo using this app, it encrypts the, you know, hashes the GPS coordinates where the photo was taken, the time, the day, and uploads that with the pixels, with the image, to Azure. And then later, when a journalist goes to use that image, they can see that the pixels are exactly the same, and then they can check the location and they can confirm the GPS. And this actually meets evidentiary standards for the UN human rights tribunal, right.BADANES: Right.DAEPP: So this is being used in Ukraine to document war crimes. And so, you know, what if everybody had that app on their phone? That means you dont you know, most photos you take, you can use an AI tool and immediately play with. But in that particular situation where you need to confirm provenance and you need to confirm that this was a real event that happened, that is a technology that exists, and I think folks like the C2PA coalition (Coalition for Content Provenance and Authenticity) can make that happen across hardware providers.NESS: And I think the challenge for me is, we cant separate this problem from some of the other, kind of, fundamental problems that we have in our media environment now, right. So, for example, if I go on to my favorite social media app and I see videos from some conflicts around the world, and these videos could be not AI generated and I still could be, you know, the target of some PR campaign to promote certain content and suppress other ones. The videos could be authentic videos, but not actually be accurate depictions of what they claim to be. And so I think that this is a the AI presents a complicating factor in an already difficult problem space. And I think, you know, trying to isolate these different variables and targeting them individually is pretty tricky. I do think that despite the liars dividend that media literacy is a very positive area to, kind of, focus energy BADANES: Yeah.NESS: in the sense that, you know, you mentioned earlier, like, using this term fraud, again, going back to this analogy with cybersecurity and cybercrime, that it tends to resonate with people. We saw that, as well, especially in Taiwan, didnt we, Madeleine? Well, in India, too, with the sextortion fears. But in Taiwan, a lot of just cybercrime in terms of defrauding people of money. And one of the things that we had observed there was that talking about generative AI in the context of elections was difficult to talk to people about it because people, kind of, immediately went into their political camps, right.BADANES: Yeah.NESS: And so you had to, kind of, penetrate you know, people were trying to, kind of, suss out which side you were on when youre trying to educate them about this topic.BADANES: Sure.NESS: But if you talk tobut everybodys, like, fraud itself is a lot less partisan.BADANES: Yeah, its a neutral term.NESS: Exactly. And so it becomes a very useful way to, kind of, get these ideas out there.BADANES: Thats really interesting. And I love the provenance example because it really gets to the question about authenticity. Like, where did something come from? What is the origin of that media? Where has it traveled over time? And if AI is a component of it, then thats a noted fact. But it doesnt put us into the space of AI or not AI, which I think is where a lot of the, sort of, labeling has gone so far. And I understand the instinct to do that. But I like the idea of moving more towards how do you know more about an image of which whether there was AI involved or not is a component but does not have judgment. That does not make the picture good or bad. It doesnt make it true or false. Its just more information for you to consume. And then, of course, the media literacy piece, people need to know to look for those indicators and want them and ask for them from the technology company. So I think thats a good, thats a good silver lining. You gave me the light at the end of the tunnel I think I was looking for on the post-truth world. So, look, heres the big question. You guys have been spending this time focusing on AI and democracy in this big, massive global election year. There was a lot of hype. [LAUGHS] There was a lot of hype. Lots of articles written about how this was going to be the AI election apocalypse. What say you? Was it? Was it not?NESS: I think it was, well, we definitely have documented cases where this happened. And Im wary of this question, particularly again from the cybersecurity standpoint, which is if you were not the victim of a terrible hack that brought down your entire company, would you say, like, well, it didnt happen, so its not going to happen, right. You would never BADANES: Yeah.NESS: That would be a silly attitude to have, right. And also, you dont know what you dont know, right. So, like, a lot of the, you know, we mentioned sextortion; we mentioned these cybercrimes. A lot of these are small-dollar crimes, which means they dont get reported or they dont get reported for reasons of shame. And so we dont even have numbers on a lot of that. And we know that the political techniques are going to mirror the criminal techniques.BADANES: Yeah.NESS: And also, I worry about, say, down-ballot elections. Like, so much of, kind of, our election this year, a lot of the focus was on the national candidates, but, you know, if local poll workers are being targeted, if disinformation campaigns are being put out about local candidates, its not going get the kind of play in the national media such that you and I might hear about it. And so Im, you know, so Ill hand it off to Madeleine, but yeah.DAEPP: So absolutely agree with Roberts point, right. If your child was affected by sextortion, if you are a country that had an audio clip go viral, this was the deepfake deluge for you, right. That said, something that happened, you know, in India as in the United States, there were major prosecutions very early on, right.BADANES: Yeah.DAEPP: So in India, there was a video. It turned out not to be a deepfake. It turned out to be a cheap fake, to your point about, you know, the question isnt whether theres AI involved; the question is whether this is an attempt to defraud. And five people were charged for this video.BADANES: Yeah.DAEPP: And in the United States, right, those Biden robocalls using Bidens voice to tell folks not to vote, like, that led to a million-dollar fine, I think, for the telecoms and $6 million for the consultant who created that. And when we talk to people in India, you know, people who work in this space, they said, well, Im not going to do that; like, Im going to focus on other things. So internal actors pay attention to these things. That really changes what people do and how they do it. And so that, I do think the work that your team did, right, to educate candidates about looking out for the stuff, the work that the MTAC (Microsoft Threat Analysis Center) did to track usage and report it, all of that, I think, was, actually, those interventions, I think, worked. I think they were really important, and I do think that what we are this absence of a deluge is actually a huge number of people making a very concerted effort to prevent it from happening.BADANES: Thats encouraging.NESS: Madeleine, you made a really important point that this deterrence from prosecution, its effective for internal actors, BADANES: Yeah.DAEPP: Yeah, thats right.NESS: right. So for foreign states who are trying to interfere with other peoples elections, the fear of prosecution is not going to be as much of a deterrent.BADANES: That is true. I will say what we saw in this election cycle, in particular in the US, was a concerted effort by the intelligence community to call out and name nation-state actors who were either doing cyberattacks or influence operations, specific videos that they identified, whether there was AI involved or not. I think that level of communication with the public while maybe doesnt lead to those actors going to jailmaybe somedaybut does in fact lead to a more aware public and therefore hopefully a less effective campaign. If people on the other end and its a little bit into the literacy space, and its something that weve seen government again in this last cycle do very effectively, to name and shame essentially when they see these things in part, though, to make sure voters are aware of whats happening. Were not quite through this big global election year; we have a couple more elections before we really hit the end of the year, but its winding down. What is next for you all? Are you all going to continue this work? Are you going build on it? What comes next?DAEPP: So our research in India actually wasnt focused specifically on elections. It was about AI and digital communications.BADANES: Ahh.DAEPP: Because, you know, again, like India is this laboratory.BADANES: Sure.DAEPP: And I think what we learned from that work is that, you know, this is going to be a part of our digital communications and our information system going forward without question. And the question is just, like, what are the viable business models, right? What are the applications that work? And again, that comes back to making sure that whatever AI you know, people when they build AI into their entire, you know, newsletter-writing system, when they build it into their content production, that they can feel confident that its safe and that it meets their needs and that theyre protected when they use it. And similarly, like, what are those applications that really work, and how do you empower those lead users while mitigating those harms and supporting civil society and mitigating those harms? I think thats an incredible, like, thatsas a researcherthats, you know, thats a career, right.BADANES: Yeah.DAEPP: Thats a wonderful research space. And so I think understanding how to support AI that is safe, that enables people globally to have self-determination in how models represent them, and that is usable and powerful, I think thats broadly BADANES: Where this goes.DAEPP: what I want to drive.BADANES: Robert, how about you?NESS: You know, so I mentioned earlier on these AI alignment issues.BADANES: Yeah.NESS: And I was really fascinated by how local and contextual those issues really are. So to give an example from Taiwan, we train these models on training data that we find from the internet. Well, when it comes to, say, Mandarin Chinese, you can imagine the proportion of content, of just the quantity of content, on the internet that comes from China is a lot more than the quantity that comes from Taiwan. And of course, whats politically correct in China is different from whats politically correct in Taiwan. And so when we were talking to Taiwanese, a lot of people had these concerns about, you know, having these large language models that reflected Taiwanese values. We heard the same thing in India about just people on different sides of the political spectrum and, kind of, looking at a YouTuber in India had walked us through this how, for example, a founding father of India, there was a disparate literature in favor of this person and some more critical of this person, and he had spent time trying to suss out whether GPT-4 was on one side or the other.BADANES: Oh. Whose side are you on? [LAUGHS]NESS: Right, and so I think for our alignment research at Microsoft Research, this becomes the beginning of, kind of, a very fruitful way of engaging with local stakeholders and making sure that we can reflect these concerns in the models that we develop and deploy.BADANES: Yeah. Well, first, I just want to thank you guys for all the work youve done. This is amazing. Weve really enjoyed partnering with you. Ive loved learning about the research and the efforts, and Im excited to see what you do next. I always want to end these kinds of conversations on a more positive note, because weve talked a lot about the weaponization of AI and, you know, how ethical areas that are confusing and but I am sure at some point in your work, you came across really positive use cases of AI when it comes to democracy, or at least I hope you have. [LAUGHS] Do you have any examples or can you leave us with something about where you see either it going or actively being used in a way to really strengthen democratic processes or systems?DAEPP: Yeah, I mean, there is just a big paper in Science, right, which, as researchers, when something comes out in Science, you know your field is about to change, right, BADANES: Yeah.DAEPP: showing that an AI model in, like, political deliberations, small groups of UK residents talking about difficult topics like Brexit, you know, climate crisis, difficult topics, that in these conversations, an AI moderator created, like, consensus statements that represented the majority opinion, still showed the minority opinion, but that participants preferred to a human-written statement and in fact preferred to their original opinion.BADANES: Wow.DAEPP: And that this, you know, not only works in these randomized controlled trials but actually works in a real citizens deliberation. And so that potential of, like, carefully fine-tuned, like, carefully aligned AI to actually help people find points of agreement, thats a really exciting space.BADANES: So next time my kids are in a fight, Im going to point them to Copilot and say, work with Copilot to mediate. [LAUGHS] No, thats really, really interesting. Robert, how about you?NESS: She, kind of, stole my example. [LAUGHTER] But Ill take it from a different perspective. So, yes, like how these technologies can enable people to collaborate and ideally, I think, from a democratic standpoint, at a local level, right. So, I mean, I think so much of our politics were, kind of, focused at the national-level campaign, but our opportunity to collaborate is much more were much more easily we can collaborate much more easily with people who are in our local constituencies. And I think to myself about, kind of, like, the decline particularly of local newspapers, local media.BADANES: Right.NESS: And so I wonder, you know, can these technologies help address that problem in terms of just, kind of, information about, say, your local community, as well as local politicians. And, yeah, and to Madeleines point, so Madeleine started the conversation talking about her background in urban planning and some of the work she did, you know, working on a local level with local officials to bring technology to the level of cities. And I think, like, well, you know, politics are local, right. So, you know, I think that thats where theres a lot of opportunity for improvement.BADANES: Well, Robert, you just queued up a topic for a whole other podcast because our team also does a lot of work around journalism, and I will say we have seen that AI at the local level with local news is really a powerful tool that were starting to see a lot of appetite and interest for in order to overcome some of the hurdles they face right now in that industry when it comes to capacity, financing, you know, not able to be in all of the places they want to be at once to make sure that theyre reporting equally across the community. This is, like, a perfect use case for AI, and were starting to see folks who are really using it. So maybe well come back and do this again another time on that topic. But I just want to thank you both, Madeleine and Robert, for joining us today and sharing your insights. This was really a fascinating conversation. I know I learned a lot. I hope that our listeners learned a lot, as well.[MUSIC]And, listeners, I hope that you tune in for more episodes of Ideas, where we continue to explore the technologies shaping our future and the big ideas behind them. Thank you, guys, so much.DAEPP: Thank you.NESS: Thank you.[MUSIC FADES][1] The video generation model Sora was released publicly earlier this month (opens in new tab).[2] For a summary of and link to the report, see the Microsoft On the Issues blog post China tests US voter fault lines and ramps AI content to boost its geopolitical interests (opens in new tab).

In our increasingly complex digital landscape, enterprises and cloud providers face significant challenges in the development, deployment, and maintenance of sophisticated IT applications. The broad adoption of microservices and cloud-based serverless architecture has streamlined certain aspects of application development while simultaneously introducing a host of operational difficulties, particularly in fault diagnosis and mitigation. These complexities can result in outages, which have the potential to cause major business disruptions, underscoring the critical need for robust solutions that ensure high availability and reliability in cloud services. As the expectation for five-nines availability grows, organizations must navigate the intricate web of operational demands to maintain customer satisfaction and business continuity.To tackle these challenges, recent research on using AIOps agents for cloud operationssuch as AI agents for incident root cause analysis (RCA) or triaginghas relied on proprietary services and datasets. Other prior works use frameworks specific to the solutions that they are building, or ad hoc and static benchmarks and metrics that fail to capture the dynamic nature of real-world cloud services. Users developing agents for cloud operations tasks with Azure AI Agent Service can evaluate and improve them using AIOpsLab. Furthermore, current approaches do not agree on standard metrics or a standard taxonomy for operational tasks. This calls for a standardized and principled research framework for building, testing, comparing, and improving AIOps agents. The framework should allow agents to interact with realistic service operation tasks in a reproducible manner. It must be flexible in extending to new applications, workloads, and faults. Importantly, it should go beyond just evaluating the AI agents and enabling users to improve the agents themselves; for example, by providing sufficient observability and even serving as a training environment (gym) to generate samples to learn on.We developed AIOpsLab, a holistic evaluation framework for researchers and developers, to enable the design, development, evaluation, and enhancement of AIOps agents, which also serves the purpose of reproducible, standardized, interoperable, and scalable benchmarks. AIOpsLab is open sourced at GitHub (opens in new tab) with the MIT license, so that researchers and engineers can leverage it to evaluate AIOps agents at scale. The AIOpsLab research paper has been accepted at SoCC24 (the annual ACM Symposium on Cloud Computing).Figure 1. System architecture of AIOpsLab.Agent-cloud interface (ACI)AIOpsLab strictly separates the agent and the application service using an intermediate orchestrator. It provides several interfaces for other system parts to integrate and extend. First, it establishes a session with an agent to share information about benchmark problems: (1) the problem description, (2) instructions (e.g., response format), and (3) available APIs to call as actions.The APIs are a set of documented tools, e.g., get logs, get metrics, and exec shell, designed to help the agent solve a task. There are no restrictions on the agents implementation; the orchestrator poses problems and polls it for the next action to perform given the previous result. Each action must be a valid API call, which the orchestrator validates and carries out. The orchestrator has privileged access to the deployment and can take arbitrary actions (e.g., scale-up, redeploy) using appropriate tools (e.g., helm, kubectl) to resolve problems on behalf of the agent. Lastly, the orchestrator calls workload and fault generators to create service disruptions, which serve as live benchmark problems. AIOpsLab provides additional APIs to extend to new services and generators.Example shows how to onboard an agent to AIOpsLabfrom aiopslab import Orchestratorclass Agent: def __init__(self, prob, instructs, apis): self.prompt = self.set_prompt(prob, instructs, apis) self.llm = GPT4() async def get_action(self, state: str) -> str: return self.llm.generate(self.prompt + state)#initialize the orchestratororch = Orchestrator()pid = "misconfig_app_hotel_res-mitigation-1"prob_desc, instructs, apis = orch.init_problem(pid)#register and evaluate the agentagent = Agent(prob_desc, instructs, apis)orch.register_agent(agent, name="myAgent")asyncio.run(orch.start_problem(max_steps=10))ServiceAIOpsLab abstracts a diverse set of services to reflect the variance in production environments. This includes live, running services that are implemented using various architectural principles, including microservices, serverless, and monolithic.We also leverage open-sourced application suites such as DeathStarBench as they provide artifacts, like source code and commit history, along with run-time telemetry. Adding tools like BluePrint can help AIOpsLab scale to other academic and production services.The workload generator in AIOpsLab plays a crucial role by creating simulations of both faulty and normal scenarios. It receives specifications from the orchestrator, such as the task, desired effects, scale, and duration. The generator can use a model trained on real production traces to generate workloads that align with these specifications. Faulty scenarios may simulate conditions like resource exhaustion, exploit edge cases, or trigger cascading failures, inspired by real incidents. Normal scenarios mimic typical production patterns, such as daily activity cycles and multi-user interactions. When various characteristics (e.g., service calls, user distribution, arrival times) can lead to the desired effect, multiple workloads can be stored in the problem cache for use by the orchestrator. In coordination with the fault generator, the workload generator can also create complex fault scenarios with workloads.Fault generatorAIOpsLab has a novel push-button fault generator designed for generic applicability across various cloud scenarios. Our approach integrates application and domain knowledge to create adaptable policies and oracles compatible with AIOps scenarios. This includes fine-grained fault injection capable of simulating complex failures inspired by production incidents. Additionally, it can inject faults at various system levels, exposing root causes while maintaining semantic integrity and considering interdependencies between cloud microservices. The fault injectors versatility can enhance the reliability and robustness of cloud systems by enabling thorough testing and evaluation of AIOps capabilities.Microsoft Research BlogMicrosoft Research Forum Episode 3: Globally inclusive and equitable AI, new use cases for AI, and moreIn the latest episode of Microsoft Research Forum, researchers explored the importance of globally inclusive and equitable AI, shared updates on AutoGen and MatterGen, presented novel use cases for AI, including industrial applications and the potential of multimodal models to improve assistive technologies.Read moreOpens in a new tab ObservabilityAIOpsLab is equipped with an extensible observability layer designed to provide comprehensive monitoring capabilities across various system layers for any AIOps tool. AIOpsLab collects a wide array of telemetry data, including (1) traces from Jaeger detailing the end-to-end paths of requests through distributed systems, (2) application logs formatted and recorded by Filebeat and Logstash, and (3) system metrics monitored by Prometheus. Additionally, AIOpsLab also captures lower-level system information such as syscall logs and cluster information. As mentioned, we handle potential data overload through flexible APIs to tune the telemetry data relevant to the AIOps tools.AIOpsLab currently supports four key tasks within the AIOps domain. These tasks include incident detection, localization, root cause diagnosis, and mitigation. Additionally, we support several popular agent frameworks: React, Autogen, and TaskWeaver. Two key insights from the study emphasize the importance of observability and well-designed ACI. Observability is crucial for clear root-cause diagnosis, for instance, pinpointing a misconfigured API gateway can be vital in preventing service downtime.Flexibility is another critical factor; the ability to execute arbitrary shell commands allowed for effective troubleshooting in real-time scenarios. Lastly, robust error handling is essentialproviding agents with high-quality feedback on execution barriers, such as a failed database connection, ensures swift resolution and continuous improvement.Example of ACI on diagnosis taskAgent: I need to first look at the logs to understand if there are any errors or warnings that could indicate problems. Action: ``` get_logs("compose-post-service", "test-social-network") ```Service: [2024-Aug-04 23:18:49.365494] <info>: Thrift: Sun Aug 4 23:19:19 2024 TSocket::open() connect() <Host: user-service Port: 9090>: Connection refused Agent: unable to connect to the `user-service` at port 9090, even though the pod is running. This suggests a possible network issue or misconfiguration in service discoveryNext stepsThis research project adopts Microsoft security standards and Responsible AI principles, and we envision this research evolving into a vital resource for organizations aiming to optimize their IT operations. Additionally, we plan to collaborate with various generative AI teams to incorporate AIOpsLab as a benchmark scenario for evaluating state-of-the-art models. By doing so, we aim to foster innovation and encourage the development of more advanced AIOps solutions. This research is essential not only for IT professionals but also for anyone invested in the future of technology, as it has the potential to redefine how organizations manage operations, respond to incidents, and ultimately serve their customers in an increasingly automated world.AcknowledgementsWe would like to thank Yinfang Chen, Manish Shetty, Yogesh Simmhan, Xuchao Zhang, Jonathan Mace, Dax Vandevoorde, Pedro Las-Casas, Shachee Mishra Gupta, and Suman Nath, for contributing to this project.Opens in a new tab