Saudi Arabia’s artificial intelligenceambitions were boosted following US president Donald Trump’s recent trip to the Middle East. The Trump administration’s import tariffs and AI chip export restrictions appear to have been behind the decision by Saudi Arabian datacentre operator DataVolt to sign a major server deal with US firm Supermicro. The company’s selection of the US firm for its bn investment in AI datacentre technology will secure graphics processing unithardware currently seen as the best way to provide the accelerated processing power AI workloads require. “Partnering with Supermicro guarantees us a US-made supply chain for critical GPU systems and positions DataVolt to accelerate our investment plans,” said Rajit Nanda, CEO of DataVolt. Supermicro said the collaboration would fast-track delivery of its ultra-dense GPU platforms, storage and rack-based systems for DataVolt’s hyperscale gigawatt-class renewable and net-zero green AI datacentre facilities.  The hardware uses liquid cooling, which Supermicro said reduces power costs by up to 40%, which allows datacentres to run more efficiently with lower power usage effectiveness. “We are excited to collaborate with DataVolt to bring our advanced AI systems featuring the latest direct liquid cooling technology, powered by local renewable, sustainable and net-zero green technology,” said Charles Liang, president and CEO of Supermicro. Supermicro claims its latest AI infrastructure is able to lower the total cost of ownership by up to 20%. Along with its Saudi AI datacentre build, Datavolt, Google, Oracle, Salesforce, AMD and Uber have said they are committed to the investment of bn in cutting-edge transformative technologies in both the US and Saudi Arabia. Oracle said it plans to invest bn over the next 10 years to deliver cloud and AI technology to Saudi Arabia. “Thanks to the decisive actions and strong leadership of President Trump and his administration, Oracle is providing the world’s most advanced cloud and AI technology to Saudi Arabia,” said Safra Catz, CEO at Oracle. “Our expanded partnership with the kingdom will create new opportunities for its economy and deliver better health outcomes for its people.” As part of Trump’s visit, Humain, a subsidiary of Saudi Arabia’s Public Investment Fund, also announced a bn deal with US chipmaker AMD to develop 500 megawatts of open, scalable, resilient and cost-efficient AI infrastructure that spans Saudi Arabia and the US. Through the collaboration, Humain will oversee end-to-end delivery, including hyperscale datacentres, sustainable power systems and global fibre interconnects, while AMD has said it will be providing its AI compute processors and the AMD ROCm open software ecosystem. Tareq Amin, CEO of Humain, described the initiative as “an open collaboration to the world’s innovators”. “We are democratising AI at the compute level, ensuring that access to advanced AI is limited only by imagination, not by infrastructure,” he said. about AI in Saudi Arabia Saudi puts bn into AI as experts debate next steps: The kingdom’s Leap 2025 tech show is the backdrop for huge investment, plus debate over the future of artificial intelligence as a productivity tool but which can also potentially undermine human society. Saudi Arabia struggling to reach global leadership in deeptech: Petrostate monarchy trying to build surrogate industry made of foreign startups because own ecosystem is too immature. A number of US tech firms have made major investments in Saudi this year, driven by the kingdom’s ambitions to boost AI. Data for the Saudi Data and AI Authority from Accenture shows that generative AIhas the potential to elevate its gross domestic productby approximately bn by augmenting and automating nearly a third of all jobs. In February, Accenture said it was collaborating with Google Cloud to accelerate the adoption of cloud and GenAI capabilities in Saudi Arabia to support the KSA’s local data, operational and software sovereignty needs. As part of this work, Google and Accenture announced a joint centre of excellence for GenAI. During the Leap 2025 conference, Salesforce said it would be investing m in Saudi, with the focus again on AI. The company announced a regional head office in Riyadh, and as part of this expansion, Salesforce said it was working with Amazon Web Servicesto deliver its Hyperforce, next-generation platform architecture to the kingdom. The work with AWS means Salesforce’s global customers are able to run workloads locally through a distributed public cloud infrastructure that Salesforce said complies with local regulations. Commenting on Salesforce’s plans, Abdullah Alswaha, minister of communications and information technology, said: “We look forward to seeing Salesforce expand its presence here and welcome the investment in AI that will drive unprecedented innovation and operational efficiency, supporting the realisation of Saudi Arabia’s Vision 2030 goals.” While the White House has linked Trump’s visit to the investments major tech firms are making in Saudi Arabia, the kingdom faces a number of challenges as it fleshes out its AI ambitions. According to a March 2025 paper from the Carnegie Endowment for International Peace, Vision 2030 represents a future-orientated vision to make religion less of a focal point of its society, particularly with regards to how Saudi Arabia is perceived internationally. But the paper notes that private-sector employers face the challenge of hiring Saudi citizens to fill high-skilled jobs. “This skills gap will only worsen as the kingdom targets a wide range of highly skilled sectors, including AI startups, green technology, datacentres, e-gaming, fintech and EV production,” it says. #trump #visit #bolsters #saudi

Foxconn has made a billion investment into its India arm, a move that will almost certainly fund its expansion of iPhone production in the country.A Foxconn factory signIndia is growing as a production base for iPhones, as Apple diversifies its supply chain away from a China-centric model. Expanding in India requires resources, and one supply chain member is contributing more.Hon Hai, also known as Foxconn, is investing billion in its India business. In a stock exchange filing seen by Bloomberg on Monday, the investment is being made via the company's subsidiary in Singapore. Continue Reading on AppleInsider | Discuss on our Forums #india #iphone #production #will #boosted

Author: Sophia Banton Originally published on Towards AI. Steve Jobs once said, “Technology is nothing. What’s important is that you have a faith in people, that they’re basically good and smart, and if you give them tools, they’ll do wonderful things with them.” These were the tools that were given to me so that I could fly, by the giants on whose shoulders I stand. PyMOL: Seeing Beauty in Science I remember it like it was yesterday. I was working on an assignment in class. Following the steps carefully, I watched as it happened: proteins appeared as beautiful ribbons on the screen, their intricate structures swirling in vibrant colors. In that moment, I was captivated by PyMOL, a computer program for viewing biological molecules in 3D. Warren Delano’s PyMOL wasn’t just a visualization tool — it was a window into the elegance of science. PyMOL taught me that data is more than information — it’s art, and that technology and science are deeply intertwined. It was also my first interaction with open-source software — free tools that bring opportunities to anyone, anywhere. This insight, the power of accessible technology, has endured among my fundamental beliefs. With PyMOL, I found the gateway to the next chapter of my journey. An image I created with PyMOL was central to my first scientific publication. That image remains on the opening page of my portfolio today, a testament to the power of visualization. That publication led to my first professional role in science, where I discovered the tool that would open the door to endless possibilities. R: Freedom to Create with Code In that role I discovered R, a programming language for graphics and statistics — it was love at first byte. Unlike PyMOL, R was my first self-taught adventure, mastered at home with just an Amazon-bought book and determination. While other programming languages felt like strict rule books, R was an artist’s palette. Its quirky symbols and flexible approach felt like an invitation to be creative with code. R became my key to exploring data, ultimately unlocking the most impactful opportunity on my path. The data manipulation skills I developed in R led me to the frontiers of innovation — a new role in biomedical research. R wasn’t just a tool; it became a trusted companion for weaving together complex data — from genomics to clinical information. With R, data analysis was just the beginning. The next tool allowed me to mesmerize audiences with the beauty of data. ggplot2: Turning Data into Colorful Stories Like my discovery of PyMOL, my first encounter with Hadley Wickham’s ggplot2 resonated deeply. This visualization toolkit for R, built on the principles of the grammar of graphics, transcended data into stories told through colors, patterns, and shapes. I wasn’t just analyzing data anymore; I was uncovering hidden stories. These plots had elements of style that would impress Van Gogh — themes, borders, and vibrant palettes. The result? Multiple scientific publications and a new identity: “the woman who makes pretty plots”. But like PyMOL and R, ggplot2 taught me that success isn’t just about achievements — it’s about empowering others. Inspired by the open-source community, I created an online ggplot2 course. The most rewarding moment? When a colleague from another continent recognized me from my course and warmly shook my hand. Yet ggplot2 wasn’t the final chapter — it was another stepping-stone on my road of discovery. Plotly: Making Data Come Alive ggplot2 revealed the beauty of data, but Plotly in R taught me how to make visualizations interactive with clickable charts and dynamic features. Visualizations were no longer just static images on screens — they could come alive. Plotly also allowed me to fine-tune my skills in another programming language called Python. Plotly in Python opened doors to freelance opportunities in data visualization. These projects boosted both my skills and confidence. These experiences prepared me for my leap into industry, where I would turn tools into solutions. But before that transition, there was one more tool in R to master — it would become my most trusted companion. R Shiny: The Catalyst for Transformation R had become the backbone of my career when I stumbled upon something unexpected — R Shiny, a tool for creating web apps in R. I stared at the screen in awe, remembering the first time I saw protein ribbons in PyMOL. I used online resources to teach myself R Shiny. R Shiny brought everything together: R’s analytics, ggplot2’s beauty, and Plotly’s interactivity. Now I could share data through intuitive web apps, no more creating endless PowerPoint presentations. Shiny became my treasured companion and the cornerstone of my budding career. R Shiny wasn’t just a tool — it was a career catalyst. Making apps wasn’t part of my original plan — honestly, there was no plan. But learning R Shiny gave me the confidence to tackle new challenges beyond the academic environment I called home. Shiny in Action: Empowering Users and Solving Problems I joined a startup where I used Shiny to detect fraud — my first venture beyond academia into the world of technology professionals. Then came an opportunity that would tie all my tools together. Still new to industry, I was unfamiliar with recruiters, hiring practices and corporate culture. But I did what I had always done, I used my best tools. The hiring process required a hands-on use case, so I built a Shiny app in two intense days. That app got me the job. Within this new role, R Shiny gave me my first industry publication and first published app. Like PyMOL opened the door to science, R Shiny introduced me to the complexities of working in industry. Each new app connected me with different business functions — from Marketing to Medical Affairs — teaching me about collaboration, resilience, and servant leadership. These experiences prepared me for an unexpected shift — the rise of AI that transformed how we interact with technology. Generative AI: Redefining Interaction The release of ChatGPT marked a turning point in how people interacted with technology. I turned to my trusted friend — R Shiny — to quickly build examples of what this new technology could do. Within two months of the release of ChatGPT, we had our first generative AI application running. Once again, R Shiny proved to be an invaluable tool for embracing the future. By the next year, generative AI had infiltrated industries, creating new opportunities for innovation. At work, I had the chance to contribute to an exciting generative AI project. The increasing demands for flexibility led me to transition to Shiny for Python, combining Shiny’s elegance with Python’s vast AI resources. The application proved successful enough to move from a prototype to an operational solution within the company. Shiny had evolved, and so had I. No longer just “the woman who makes pretty plots and apps”, I stepped into the future of AI with my trusted companion at my side. Because regardless of the engines that power AI, the need to make data accessible and interactive will always remain. My Tools, My Teammates Looking back, these weren’t just tools — they were teammates. PyMOL revealed the beauty of science. R offered boundless creativity. ggplot2 and Plotly turned data into stories. Shiny transformed me from a scientist to an innovator, ready for the AI revolution. Each tool shaped who I am, and together they taught me the most important lesson: technology’s true power lies not in the code, but in how it empowers people to do wonderful things. About the Author Sophia Banton is an Associate Director and AI Solution Lead in biopharma, specializing in Responsible AI governance, workplace AI adoption, and building and scaling AI solutions across IT and business functions. With a background in bioinformatics, public health, and data science, she brings an interdisciplinary lens to AI implementation — balancing technical execution, ethical design, and business alignment in highly regulated environments. Her writing explores the real-world impact of AI beyond theory, helping organizations adopt AI responsibly and sustainably. Connect with her on LinkedIn or explore more AI insights on Medium. Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor. Published via Towards AI #journey #tools #that #opened #each

Michael M. Santiago/Getty Images; Getty Images; BIWall Street banks are proving that generative AI is here to stay, and the tech is not just a fad.Business Insider has reported on how some of finance's biggest banks are approaching generative AI.See how giants like Goldman Sachs and JPMorgan are weaving the tech into the fabric of their firms.Wall Street bank leaders say generative AI is here to stay, and they're weaving the technology throughout the fabric of their banks to make sure.From trading to payments to marketing, it's hard to find a corner of the banking industry that isn't claiming to use AI.In fact, the technology's impact, made mainstream by OpenAI's ChatGPT in late 2022, is becoming cultural. Generative AI is changing what it takes to be a software developer and how to stand out as a junior banker, especially as banks mull over how to roll out autonomous AI agents. The technology is even changing roles in the c-suite. But it's also presented new challenges — bank leaders say they are struggling to keep up with AI-powered cyberattacks.From supercharging productivity via AI-boosted search engines to figuring out the best way banks can realize a return on their AI investments, here's what we know about how Wall Street banks are embracing AI.JPMorgan ChaseJPMorgan CEO Jamie DimonTom Williams/CQ-Roll Call, Inc via Getty ImagesJPMorgan CEO Jamie Dimon is a "tremendous" user of the bank's generative AI suite. We have the story of how he and other bank executives use AI.JPMorgan's AI rollout: Jamie Dimon's a 'tremendous' user and it's caused some 'healthy competition' among teamsJPMorgan is making a big bet on AI. Here's how its private bankers are using it.Dimon also laid out his vision for how America's largest bank will win the AI battle against fintechs through data. Meet the leaders of that mission.Jamie Dimon says he's out to win the AI arms race. See who he's put in charge of this critical mission.JPMorgan's Lori Beer just added 2 new tech leaders to her ranks. Meet the team behind the bank's AI future.Mary Erdoes, the boss of JPM's asset- and wealth-management business, used these slides to outline how she wants to get her people ready for the "AI of the future."Here's how JPM boss Mary Erdoes wants to use AI to eliminate "no joy" workIt's not just JPMorgan's in-house tech teams that have been gearing up for an AI future. Cloud partners, like AWS, also play an important role.AWS is helping financial giants like JPMorgan and Bridgewater with their AI ambitionsGoldman SachsGoldman Sachs' David SolomonMichael KovacIs Goldman in its AI era? These real-world stories about employees using AImake it seem so. Take a look at how AI is being put to the test across the bank and seniority levels, from C-suites to analysts.7 Goldman Sachs insiders explain how the bank's new AI sidekick is helping them crush it at workGoldman is assembling a growing arsenal of AI tools. Here's everything we know about 5.AI is doing 95% of the work on an IPO prospectus, Goldman Sachs CEO saysGoldman's top partners and CEO David Solomon are eager to see AI rev up their businesses. From realizing internal productivity gains to capturing more business as clients look to raise money in anticipation of AI development and acquisitions, here's what the top echelon is expecting.How AI will shake up Goldman Sachs, according to top partners and CEO David SolomonInside Goldman Sachs' plans for AI, from helping non-tech workers do more with software to streamlining how code is documentedThere is no AI without data, and there is no data strategy at Goldman without its chief data officer, Neema Raphael. Raphael gave BI an inside look at how his roughly 500-person team melds with the rest of the bank to get the most out of its data.Meet Neema Raphael, the data whiz key to Goldman's AI ambitions who's overseeing the bank's army of engineers and scientistsAI's impact has ripple effects that go far beyond technology. Goldman's chief information officer, Marco Argenti, predicts that cultural change will be critical to getting the bank to 100% adoption.Goldman Sachs CIO Marco Argenti says Wall Street's 'next big wave' in AI will come down to culture, not just technologyMany dollars are being spent on Wall Street's AI ambitions. But how do you measure the return on the investment? Argenti offers some tips on the calculus that can help firms prioritize where to invest.As AI usage ramps up on Wall Street, Goldman Sachs CIO details how the bank analyzes return on investment for the techMorgan StanleyMorgan Stanley CEO Ted PickJeenah Moon / ReutersMorgan Stanley wants to turn employees' AI ideas into a reality. Here's an exclusive look at that process.Morgan Stanley has 30 AI projects in the pipeline. Here's how the bank sources employees' ideas for inspiration.See how AI is transforming Morgan Stanley's wealth division and the jobs of its 16,000 financial advisors.Morgan Stanley is betting on AI to free up advisors' time to be 'more human.' Nearly 100% of advisor teams use it, and here's how.Thanks to its partnership with ChatGPT-maker OpenAI, Morgan Stanley has ramped up its AI efforts. The exec in charge of tech partnerships and firmwide innovation opened up about how it all started.Morgan Stanley's new innovation head lays out his plan for more OpenAI-type partnershipsA Morgan Stanley exec breaks down how the bank courts tech partners like OpenAICitiCiti's Jane FraserNICHOLAS KAMM/Getty ImagesMeet the new exec in charge of giving an AI facelift to Citi's lagging wealth business.Citi snags AI leader from Morgan Stanley to help turn around its wealth techCiti's top tech executive, Shadman Zafar, outlined the bank's four-phased AI strategy and how it will "change how we work for decades to come."A top Citi tech exec breaks down the 4 phases of the bank's AI strategy that will impact everyone from operations to wealth and productBank of AmericaBank of America's Brian MoynihanJohn Lamparski/Getty ImagesBank of America's chief experience officer, Rob Pascal, details how the bank's internal-facing AI assistant helps bankers collect, record, and review client data. Here are all the ways it's helping employees be more effective and efficient.How Bank of America is using an AI-powered tool to help its bankers prep for client meetings more efficientlyAI hits the investment bankWall Street investment banks prepare for an AI future.Momo Takahashi/BIInvestment bankers are hopeful that corporate America's obsession with AI could kick off a new era of mergers, acquisitions, and IPOs. From execs stepping into recently created roles to accommodate the sector to industry veterans launching their own AI-focused M&A-advisory firm, meet 11 investment bankers poised to lead Wall Street's AI revolution.Wall Street is gearing up for an AI shopping spree. Meet 11 bankers poised to come out on top.We spoke with four of those AI bankers about why 2025 is going to be all about AI pickaxes and shovels rather than pure-play AI deals.The deals 4 tech bankers think the 'AI arms race' could drive in 2025AI could save junior bankers time by automating tedious tasks known all too well by Wall Street's youngest ranks. But it can also make it harder to break into the industry by shifting the skills required for entry.How AI will change the job of the junior banker and raise the bar to entryA former Goldman Sachs managing director built an AI-powered networking tool to spur dealmaking. The budding startup, Louisa AI, already has a few clients, including Goldman Sachs, Insight Partners, and a global exchange.See the pitch deck that helped a former Goldman Sachs exec raise million for an AI-powered networking startup that feeds deal ideas to bankers.Here's how former investment bankers left their Wall Street jobs to build an AI startup to solve junior bankers' woes.See the pitch deck that helped former investment bankers raise a million seed round for their generative 'AI consigliere' to bankers and analystsRead the original article on Business Insider #here039s #everything #know #about #how

Ecovacs Deebot X9 Pro Omni robot vacuum review: superior mopping skills MSRP Score Details “The Ecovacs Deebot X9 Pro Omni features one of the best mopping systems ever used in a robot vacuum.” Pros Innovative mopping system Retractable brushes and mops Self-cleaning dock Anti-tangle vacuuming Cons Expensive Too tall to fit under furniture Ecovacs launched the Deebot X8 Pro Omni earlier this year. Featuring a unique roller mop system, it offered one of the best mopping performances we’d ever seen in a robot vacuum. Now, a few months later, the company has already launched its successor — the Ecovacs Deebot X9 Pro Omni. It takes everything we love about the X8 and makes it better, including a new mop-lifting system, enhanced vacuuming system, and a fantastic anti-tangle design that eliminates hair and string from clogging up its brushes. Recommended Videos If you’re looking for a robot vacuum to thoroughly clean hard floors, you can’t go wrong with the Ecovacs Deebot X9 Pro Omni. It’s not perfect, but aside from a few minor complaints, the expensive X9 Pro Omni should make a premium addition to any smart home. One of the best mopping systems ever Jon Bitner / Digital Trends I’ve tested dozens of robot vacuums over the years, and I can say without a doubt that the Ecovacs Deebot X9 Pro Omni offers one of the best mopping experiences on the market. Known as “OZMO Roller Mopping Technology,” it’s essentially a giant roller brush on the bottom of the robot. A series of sprayers are constantly rewetting the mop to ensure a consistent clean across your floors, while a scraper is used to pull dirty water off its surface. The end result is remarkably shiny floors. Combined with its ability to apply 3,700 Pa of pressure and roll at speeds up to 220 rpm, it eliminates just about any stain or spill it encounters. Ecovacs also outfitted the X9 Pro Omni with some cool bonus features. The first is a triple-lift system, allowing the mop, main brush, and side brush to retract depending on the task at hand. When a spill is detected, both mops will retract, ensuring they stay clean. When carpet is detected, the mop will retract, ensuring your floors stay dry. Better yet, the robot offers a “Carpet First” function — meaning it’ll zip around your home and vacuum with a dry mop before starting the wetting and mopping process. When all of this is taken together, the end result is a robot vacuum perfect for cleaning hard floors. It’s also great that it can traverse carpets with dry mops — though it would be even better if it could automatically detach its mop and leave it in the base station when vacuuming. It’s a feature some competitors are starting to use to great effect, as it ensures your mops aren’t touching carpet at all. Vacuuming and self-cleaning Jon Bitner / Digital Trends Oddly enough, the X9 Pro Omni offers less suction than the older X8 Pro Omni. The previous robot packed in 18,000 Pa — the X9 Pro Omni shifts it down to just 16,600. However, its vacuuming performance remains about the same, thanks to the inclusion of BLAST technology. BLAST is shorthand for “Boosted Large-Airflow Suction Technology,” and it works via a combination of suction and maximized airflow. The idea revolves around using additional airflow to knock loose dust particles and other debris, which can then be suctioned up without requiring as much force. In practice, we’ve found the system to work incredibly well. Along with handling dust and small particles, it picks up clumps of dirt and larger objects without much trouble. It also leaves fantastic lines on plush carpet — giving your home a “just cleaned” appearance most shoppers will appreciate. The side brush doesn’t swing out like some competitors, but it still did a great job of tackling corners and tight spaces, thanks to its impressive navigation skills. Best of all, you can use onboard AI to allow the robot to determine all the best settings for a superior cleaning. There are dozens of ways to manually customize its performance via the mobile app, but I’ve found letting the robot do its own thing often results in the best clean. Jon Bitner / Digital Trends Once a cleaning cycle is complete, the robot heads back to its dock to recharge and clean itself. The docking station washes the mop with hot water, dries it with hot air, and can even dispense a cleaning solution into the robot’s water tank for a better mopping experience. The onboard dustbin will also be siphoned off into a larger dustbag in the docking station, allowing you to go 90 days before needing a replacement. Finally, the docking station will clean itself, washing away any gunk removed from the robot during the cleaning cycle. Pretty much all of the maintenance tasks are taken care of for you. You’ll just need to monitor the clean and dirty water tanks, and perform a more thorough washing tray cleaning every 150 days. As far as robot vacuums are concerned, that’s a huge win for autonomy. Verdict There’s a lot to love about the Ecovacs Deebot X9 Pro Omni, and it’s an easy recommendation for most shoppers. If your home is largely composed of hard floors, it’s an even easier choice. However, there are a few limitations to consider. For one, it can’t leave its mop behind in its docking station. Yes, it will raise it while traveling on carpet, but most shoppers will likely find the auto-detachable mops of the Roborock Saros 10 to be a more elegant solution. The robot is also quite large. At 3.8 inches tall, it’s almost a full inch taller than some competitors. So while it’s small enough to clean under most furniture, smaller nooks and crannies will be too tight for the X9 Pro Omni to fit underneath. These are minor complaints on an otherwise stellar device. The Ecovacs Deebot X9 Pro Omni proves the brand has what it takes to stick with the leaders of the segment, offering a superior mopping experience matched by few rivals. Add in the effective BLAST vacuuming system, self-cleaning dock, and impressive onboard AI, and you’ve got one of the best robot vacuums of 2025. #ecovacs #deebot #pro #omni #robot

The growth in developing and deploying large language modelsis closely tied to architectural innovations, large-scale datasets, and hardware improvements. Models like DeepSeek-V3, GPT-4o, Claude 3.5 Sonnet, and LLaMA-3 have demonstrated how scaling enhances reasoning and dialogue capabilities. However, as their performance increases, so do computing, memory, and communication bandwidth demands, placing substantial strain on hardware. Without parallel progress in model and infrastructure co-design, these models risk becoming accessible only to organizations with massive resources. This makes optimizing training cost, inference speed, and memory efficiency a critical area of research. A core challenge is the mismatch between model size and hardware capabilities. LLM memory consumption grows over 1000% annually, while high-speed memory bandwidth increases by less than 50%. During inference, caching prior context in Key-Valuestores adds to memory strain and slows processing. Dense models activate all parameters per token, escalating computational costs, particularly for models with hundreds of billions of parameters. This results in billions of floating-point operations per token and high energy demands. Time Per Output Token, a key performance metric, also suffers, impacting user experience. These problems call for solutions beyond simply adding more hardware. Techniques like Multi-Query Attentionand Grouped-Query Attentionreduce memory usage by sharing attention weights. Windowed KV caching lowers memory use by storing only recent tokens, but can limit long-context understanding. Quantized compression with low-bit formats like 4-bit and 8-bit cuts memory further, though sometimes with trade-offs in accuracy. Precision formats such as BF16 and FP8 improve training speed and efficiency. While useful, these techniques often tackle individual issues rather than a comprehensive solution to scaling challenges. Researchers from DeepSeek-AI introduced a more integrated and efficient strategy with the development of DeepSeek-V3, designed to scale intelligently rather than excessively. Utilizing 2,048 NVIDIA H800 GPUs, the model achieves state-of-the-art performance while focusing on cost-efficiency. Instead of depending on expansive infrastructure, the team engineered the model architecture to work harmoniously with hardware constraints. Central to this effort are innovations such as Multi-head Latent Attentionfor memory optimization, a Mixture of Expertsframework for computational efficiency, and FP8 mixed-precision training to accelerate performance without sacrificing accuracy. A custom Multi-Plane Network Topology was also employed to minimize inter-device communication overhead. Collectively, these components make DeepSeek-V3 a scalable and accessible solution, capable of rivaling much larger systems while operating on significantly leaner resources. The architecture achieves memory efficiency by reducing the KV cache requirement per token to just 70 KB using MLA, compared to 327 KB and 516 KB in Qwen-2.5 and LLaMA-3.1, respectively. This reduction is accomplished by compressing attention heads into a smaller latent vector jointly trained with the model. Computational efficiency is further boosted with the MoE model, which increases total parameters to 671 billion but only activates 37 billion per token. This contrasts sharply with dense models that require full parameter activation. For example, LLaMA-3.1 needs 2,448 GFLOPS per token, while DeepSeek-V3 operates at just 250 GFLOPS. Also, the architecture integrates a Multi-Token Predictionmodule, enabling the generation of multiple tokens in a single step. The system achieves up to 1.8x improvement in generation speed, and real-world measurements show 80-90% token acceptance for speculative decoding. Using a system interconnected by CX7 400 Gbps InfiniBand NICs, DeepSeek-V3 achieves a theoretical TPOT of 14.76 milliseconds, equal to 67 tokens per second. With higher-bandwidth setups like NVIDIA GB200 NVL72 offering 900 GB/s, this number can be reduced to 0.82 milliseconds TPOT, potentially achieving 1,200 tokens per second. The practical throughput is lower due to compute-communication overlap and memory limitations, but the framework lays the foundation for future high-speed implementations. FP8 precision further adds to the speed gains. The training framework applies tile-wise 1×128 and block-wise 128×128 quantization, with less than 0.25% accuracy loss compared to BF16. These results were validated on smaller 16B and 230B parameter versions before integration into the 671B model. Several key takeaways from the research on insights into DeepSeek-V3 include: MLA compression reduces KV cache size per token from 516 KB to 70 KB, significantly lowering memory demands during inference. Only 37 billion of the 671 billion total parameters are activated per token, dramatically reducing compute and memory requirements without compromising model performance. DeepSeek-V3 requires just 250 GFLOPS per token, compared to 2,448 GFLOPS for dense models like LLaMA-3.1, highlighting its computational efficiency. Achieves up to 67 tokens per secondon a 400 Gbps InfiniBand network, with the potential to scale to 1,200 TPS using advanced interconnects like NVL72. Multi-Token Predictionimproves generation speed by 1.8×, with a token acceptance rate of 80-90%, enhancing inference throughput. FP8 mixed-precision training enables faster computation with less than 0.25% accuracy degradation, validated through extensive small-scale ablations. Capable of running on a server equipped with a consumer-grade GPU, delivering nearly 20 TPS, making high-performance LLMs more accessible. In conclusion, the research presents a well-rounded framework for building powerful and resource-conscious large-scale language models. By directly addressing fundamental constraints, such as memory limitations, high computational costs, and inference latency, the researchers demonstrate that intelligent architecture-hardware co-design can unlock high performance without relying on vast infrastructure. DeepSeek-V3 is a clear example of how efficiency and scalability coexist, enabling broader adoption of cutting-edge AI capabilities across diverse organizations. This approach shifts the narrative from scaling through brute force to scaling through smarter engineering. Check out the Paper. All credit for this research goes to the researchers of this project. Also, feel free to follow us on Twitter and don’t forget to join our 90k+ ML SubReddit. Sana HassanSana Hassan, a consulting intern at Marktechpost and dual-degree student at IIT Madras, is passionate about applying technology and AI to address real-world challenges. With a keen interest in solving practical problems, he brings a fresh perspective to the intersection of AI and real-life solutions.Sana Hassanhttps://www.marktechpost.com/author/sana-hassan/Meet LangGraph Multi-Agent Swarm: A Python Library for Creating Swarm-Style Multi-Agent Systems Using LangGraphSana Hassanhttps://www.marktechpost.com/author/sana-hassan/ByteDance Introduces Seed1.5-VL: A Vision-Language Foundation Model Designed to Advance General-Purpose Multimodal Understanding and ReasoningSana Hassanhttps://www.marktechpost.com/author/sana-hassan/Researchers from Tsinghua and ModelBest Release Ultra-FineWeb: A Trillion-Token Dataset Enhancing LLM Accuracy Across BenchmarksSana Hassanhttps://www.marktechpost.com/author/sana-hassan/Coding Agents See 75% Surge: SimilarWeb’s AI Usage Report Highlights the Sectors Winning and Losing in 2025’s Generative AI Boom #this #paper #deepseekai #explores #how

Easing trade tensions between the U.S. and China and the axing of the Biden administration's chip export rule helped the sector. #nvidia #tesla #ripped #higher #this

The AMOC is a system of ocean currents that circulates water within the Atlantic OceanNASA/Goddard Space Flight Center Scientific Visualization Studio The slowdown of a major current in the Atlantic Ocean is boosting the sea level and associated flooding in the Northeast US, on top of the already-rising sea level due to climate change. A total collapse of this Atlantic Meridional Overturning Circulationas the planet warms could raise the sea level even further. “If the AMOC collapsed, this would dramatically increase the flood frequency along the US coast, even in the absence of strong storms,” says Liping Zhang at the Geophysical Fluid Dynamics Laboratory of the US National Oceanic and Atmospheric Administrationin New Jersey. “Even partial weakeningcan already have substantial impacts.” Advertisement Melting ice sheets and warmer water due to human-caused climate change are leading to a rise in average sea level, but the rate of sea-level rise isn’t the same everywhere. For instance, some coastal land is sinking, speeding the relative rate of sea-level rise in those areas. Local sea level is also shaped by how heat, water and salt circulate in the ocean, with warmer and fresher water taking up more space than colder, saltier water. On the US north-east coast, the sea level is rising three to four times faster than the global average over recent decades. A slowdown in the AMOC – which transports warm water from lower latitudes to the North Atlantic, where it cools, gets saltier and sinks – has long been proposed as a possible driver for this. When this overturning circulation weakens, deep water along the path of the current is expected to warm and expand, sloshing more water onto the shallow continental shelf. The AMOC naturally varies in strength on different timescales, and climate change has contributed to a slowdown in recent decades as melting ice freshens the North Atlantic and its waters warm. But it wasn’t clear whether this slowdown was making a big difference to sea level. Unmissable news about our planet delivered straight to your inbox every month. Sign up to newsletter Zhang and her colleagues used tide gauge measurements along the New England coast to reconstruct the local sea level stretching back more than a century. On top of a steady rise due to climate change, they found a marked pattern of fluctuation between low and high sea levels every few decades. Years with a high sea level aligned closely with years when the AMOC was weak, and these years also had more frequent coastal flooding. The researchers then used two different ocean models to quantify how much fluctuations in the AMOC’s strength influenced the local sea level. While the main driver of changes was the steady rise due to climate change, they found the weakening AMOC substantially boosted the sea level and associated flooding. In different parts of the coast, they found that a slowdown in the AMOC was behind 20 to 50 per cent of flooding since 2005. Because the natural cycles in the AMOC’s strength are largely predictable, the findings could enable researchers to forecast which years will see lots of flooding up to three years in advance, says Zhang. This could help make long-term decisions about infrastructure and emergency preparedness. “It demonstrates that the AMOC really does matter to,” says Chris Hughes at the University of Liverpool, UK, who wasn’t involved in the research. “It’s not just there in models or theory, it’s actually there in the real world.” It isn’t clear how much of the recent weakening of the AMOC is due to climate change and how much is due to natural variations. However, the findings add support for projections that much of the US East Coast could see a surge in the sea level if the AMOC were to completely collapse in response to climate change – which, though unlikely, is possible. A near-total collapse of the current could raise the sea level by around 24 centimetres, says Hughes. “It doesn’t sound very huge, but it doesn’t have to go up very much to have a big effect.” Topics: #east #coast #faces #rising #seas

Oh it’s fun to have good news! The latest set from the Pokémon TCG, Scarlet & Violet Destined Rivals, is a top-notch collection of cards bursting with Team Rocket antics, which sportspull rates I don’t believe we’ve seen in this era. Having torn open 55 packs, I have a generous spread of rares and ex cards the likes of which I’ve not seen since S&V began. Also, unlike the awful previous set Journey Together, it’s an excellent collection of Trainer Pokémon to really charge up the live game.Suggested ReadingWhat’s Coming Out Beyond Pokémon: The Indigo Disk | The Week In Games Share SubtitlesOffEnglishview videoSuggested ReadingWhat’s Coming Out Beyond Pokémon: The Indigo Disk | The Week In Games Share SubtitlesOffEnglishYou know, if you can buy it.While it’s delightful to report that The Pokémon Company has really knocked it out of the park with Destined Rivals, unfortunately this hasn’t coincided with addressing the wild shortages of cards for regular customers. It’s a problem that only just repeated itself with last week’s website-crashing launch of the next sets, White Fire and Black Bolt—the first ever split-set English-language collection—that looks likely to be as impossible to buy as just about everything else this year. Or, if you do, you’ll be paying way over MSRP to scalpers, and please don’t do that.However, smart players will know that the best way to get cards for any set is to attend a pre-release event at their local store or club, where everyone receives seven or more packs, generally for less than Those are taking place Saturday and Sunday, May 17 and 18, although I hear that even these were booking up fast weeks ago. If you can, I really recommend making the effort for Destined Rivals. It’s a bunch of fun, and if the 55 packs I opened are an accurate sample, you’re likely to pick up a good handful of super-pretty full-art cards.Image: The Pokémon Company / KotakuSo why am I so excited for this set? It’s a combination of things. It’d be silly to pretend that the first appearance of Team Rocket in the game in 25 years isn’t a big part of the thrill, and the set is rammed full of the nefarious group and their signature monsters. Among the Pokémon boosted by Team Rocket are Moltres, Zapdos and Articuno, along with newcomer Spidops, classics like Meowth and Mewtwo, and that most evil of Pokémon, Flaaffy. Meanwhile, for the forces of good, Cynthia, Misty, Ethan and Arven all join, again bringing back some favorites like Milotic, Gyarados, Psyduck and Ho-oh. It’s a real crowd-pleaser.Secondly, those pull-rates. When I get these boxes of cards from TPCi, I sit down and open them with my 10-year-old. It’s a really solid way of gauging the levels of satisfaction, his spirit draining out of him when we’re tearing through a set like Journey Together and just getting endless bulk. But with Destined Rivals, even my Pokémon-uninterested wife wanted in, so fun was it to have a strong chance of finding an exciting card. Where Journey Together only had 31 full-art cards, Destined Rivals has an amazing 62! Double! Admittedly, that’s on top of a wild 182 regular cards, making this the biggest set since Surging Sparks, but with—in my admittedly unscientific sample—a seemingly much better chance of finding the special stuff.We were especially lucky to pull the Team Rocket’s Ariana Special Illustration Rare, along with one of my chase cards, the Illustration Rare of Misty’s Psyduck. No Mewtwo, sadly, but we also got 12 regular ex cards, and 11 full-arts! If you include ex in the figures, that’s a pull-rate of almost one in two! Remove the regular ex cards and you’ve still got one in five for something Ultra Rare or better. Those included the wildly gorgeous Rapidash by Rond, Mori Yuu’s extraordinarily detailed Clamperl, the delightful Team Rocket’s Murkrowby Akira Komayama, and the splendidly silly Team Rocket’s Raticate by Mekayu.Image: The Pokémon Company / KotakuAnd thirdly, the game itself! Journey Together was supposed to be the reintroduction of Trainer Pokémon to the live game, but it was such a damp squib. This time, things are really going to get mixed up! Team Rocket arrive with an array of brand new tricks and cheats, and while people are obviously going to build decks around Misty and Ethan, it’ll be the baddies that once more prove the most fun.There’s the addition of Team Rocket’s Energy, which provides two energy to any Team Rocket Pokémon, and can be either Dark or Psychic or both! Meanwhile, Team Rocket’s Venture Bomb lets you flip a coin to find out if it’s going to do 20 damage to any of your opponent’s Pokémon, or 20 specifically to your own Active Pokémon—but being an Item card, you can do this silliness as many times as you have cards in a single turn. Giovanni, meanwhile, offers a classic evil move: you can play him to swap out your current Active Pokémon, but also do the same to your opponent, and choose which of their benched Pokémon goes in. Team Rocket’s Great Ball lets you flip a coin and then pull either an Evolution or Basic Pokémon from your deck depending on the result. And then Stadium card Team Rocket’s Watchtower renders all Pokémon without abilities! That’s going to destroy so many players’ tactics!Oh, and there’s a card called Team Rocket’s Bother Bot, and while its ability is fascinating—you can flip one of your opponent’s prize cards, then pick a random card from their hand, and then choose if you want them to swap them over—I’m mostly mentioning it because I find its name very funny.Then the Team Rocket Pokémon themselves do some real mischief. Arbok, for instance, stops your opponent playing any card with an ability, and also does 30 damage to every single Pokémon your opponent has on the board. Articuno can prevent all attack effects just by being on the bench. Dottler lets you look at the top five cards on your opponent’s deck, and then put them back in your preferred order! Ha!Nidoran ♀ and ♂ offer their usual teamwork options, but super-powerfully. Once you’ve evolved to Nidorina, you can do an attack that lets you search your deck to evolve any two of your benched monsters, and then Stage 2's Nidoqueen will do 180 damage for one energy if you have a Nidoking in play. Oh, and Ampharos, Flaaffy’s ultimate form, has an ability that means any time your opponent evolves a Pokémon, they automatically put 40 damage on it! That’s monstrous.Image: The Pokémon Company / KotakuIt’s going to be so interesting to see how people manipulate these new additions into the meta, not least when yet more cards are deliberately designed to mess up current favorite decks—Mimikyu lets you steal Tera Pokémon attacks, for instance. I think this should finally offer the shake-up the game needs in its third year of this era, beforenext year’s switch to Mega Pokémon instead of a fourth year of S&V.Now, as I’ve said, my sample of 55 packs isn’t big enough to be indicative, and perhaps we just got weirdly lucky. But I have high hopes here. We’ll get a proper idea when the likes of Danny Phantump have put together their pull-rate data. Either way, there’s such a wealth of beautiful cards in the set, so much incredible art to collect, and a bunch that’ll make the live game so very interesting. Which is pretty much all I can ask for from a Pokémon TCG set. Other than, you know, being able to buy it. Which is going to be very, very hard to do.Destined Rivals officially releases on May 30, with pre-release events taking place this weekend, May 17-18.. #pokémon #tcg039s #return #team #rocket