OpenAI CEO Sam Altman laid out a big vision for the future of ChatGPT at an AI event hosted by VC firm Sequoia earlier this month.  When asked by one attendee about how ChatGPT can become more personalized, Altman replied that he eventually wants the model to document and remember everything in a person’s life. The ideal, he said, is a “very tiny reasoning model with a trillion tokens of context that you put your whole life into.” “This model can reason across your whole context and do it efficiently. And every conversation you’ve ever had in your life, every book you’ve ever read, every email you’ve ever read, everything you’ve ever looked at is in there, plus connected to all your data from other sources. And your life just keeps appending to the context,” he described. “Your company just does the same thing for all your company’s data,” he added. Altman may have some data-driven reason to think this is ChatGPT’s natural future. In that same discussion, when asked for cool ways young people use ChatGPT, he said, “People in college use it as an operating system.” They upload files, connect data sources, and then use “complex prompts” against that data. Additionally, with ChatGPT’s memory options — which can use previous chats and memorized facts as context — he said one trend he’s noticed is that young people “don’t really make life decisions without asking ChatGPT.”  Techcrunch event Join us at TechCrunch Sessions: AI Secure your spot for our leading AI industry event with speakers from OpenAI, Anthropic, and Cohere. For a limited time, tickets are just $292 for an entire day of expert talks, workshops, and potent networking. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 REGISTER NOW “A gross oversimplification is: Older people use ChatGPT as, like, a Google replacement,” he said. “People in their 20s and 30s use it like a life advisor.” It’s not much of a leap to see how ChatGPT could become an all-knowing AI system. Paired with the agents the Valley is currently trying to build, that’s an exciting future to think about.  Imagine your AI automatically scheduling your car’s oil changes and reminding you; planning the travel necessary for an out-of-town wedding and ordering the gift from the registry; or preordering the next volume of the book series you’ve been reading for years. But the scary part? How much should we trust a Big Tech for-profit company to know everything about our lives? These are companies that don’t always behave in model ways. Google, which began life with the motto “don’t be evil” lost a lawsuit in the U.S. that accused it of engaging in anticompetitive, monopolistic behavior.  Chatbots can be trained to respond in politically motivated ways. Not only have Chinese bots been found to comply with China’s censorship requirements but xAI’s chatbot Grok this week was randomly discussing a South African “white genocide” when people asked it completely unrelated questions. The behavior, many noted, implied intentional manipulation of its response engine at the command of its South African-born founder, Elon Musk. Last month, ChatGPT became so agreeable it was downright sycophantic. Users began sharing screenshots of the bot applauding problematic, even dangerous decisions and ideas. Altman quickly responded by promising the team had fixed the tweak that caused the problem. Even the best, most reliable models still just outright make stuff up from time to time.  So, having an all-knowing AI assistant could help our lives in ways we can only begin to see. But given Big Tech’s long history of iffy behavior, that’s also a situation ripe for misuse.

Disney has made no secret of its changing strategy toward streaming in the last year or two. When Disney+ first debuted, the company allocated an enormous amount of resources to produce movies and shows exclusively for streaming. In the last year or two, they’ve openly discussed shifting a lot of those resources back to content that can debut in theaters first, and then move to streaming.That change is evident each month when Disney unveils the new batch of content coming to Disney+. There’s just not that much new stuff in these dispatches lately, especially when compared to a couple of years ago.In June of 2025, there will be a new Marvel series — Ironheart, spinning off a young armored hero who first debuted in Black Panther: Wakanda Forever — and there’s a new season of the animated series Phineas and Ferb. There’s also a filmed version of the Frozen Broadway musical, and documentaries about oceans and astronaut Sally Ride. That’s really most of the big new additions for the entire month. (If you also subscribe to Hulu, you’ll also get access to two other big titles: an brand-new animated Predator film and Season 4 of the hit series The Bear.)Here’s the full list of what’s coming to Disney+ in June 2025...Wednesday, June 4- Pupstruction (S2, 6 episodes)PHINEAS, FERBDisneyloading...READ MORE: The Best Disney+ Movies You Might Have MissedFriday, June 6New to Disney+Phineas and Ferb (Season 5) - Premiere, First 10 EpisodesSunday, June 8New to Disney+Ocean with David Attenborough - PremiereTuesday, June 17New to Disney+SALLY - PremiereFriday, June 20New to Disney+Frozen: The Hit Broadway Musical - PremiereIRONHEARTMarvelloading...Tuesday, June 24Disney+ OriginalIronheart - Three Episode Premiere at 6pm PTComing in June- Underdogs (Season 1)Exclusively on Disney+Vibe CheckSign up for Disney+ here.Get our free mobile appThe Most Underrated Disney Movies Of The Last 25 YearsNo one considers these Disney movies “classics” — but they should.

VLMs have become central to building general-purpose AI systems capable of understanding and interacting in digital and real-world settings. By integrating visual and textual data, VLMs have driven advancements in multimodal reasoning, image editing, GUI agents, robotics, and more, influencing sectors like education and healthcare. Despite this progress, VLMs still lag behind human capabilities, particularly in tasks involving 3D reasoning, object counting, creative visual interpretation, and interactive gameplay. A challenge lies in the scarcity of rich, diverse multimodal datasets, unlike the abundant textual resources available to LLMs. Additionally, multimodal data complexity poses significant training and evaluation hurdles.  Researchers at ByteDance have developed Seed1.5-VL, a compact yet powerful vision-language foundation model featuring a 532 M-parameter vision encoder and a 20 B-parameter Mixture-of-Experts LLM. Despite its efficient architecture, Seed1.5-VL achieves top results on 38 out of 60 public VLM benchmarks, excelling in tasks like GUI control, video understanding, and visual reasoning. It is trained on trillions of multimodal tokens using advanced data synthesis and post-training techniques, including human feedback. Innovations in training, such as hybrid parallelism and vision token redistribution, optimize performance. The model’s efficiency and strong reasoning capabilities suit real-world interactive applications like chatbots.  The Seed1.5-VL architecture features a vision encoder, an MLP adapter, and an LLM. Its custom vision encoder, Seed-ViT, supports native-resolution image input using 2D RoPE and processes images through 14×14 patches, followed by average pooling and an MLP. Pretraining involves masked image modeling, contrastive learning, and omni-modal alignment using images, text, and video-audio-caption pairs. The model uses a Dynamic Frame-Resolution Sampling approach for video encoding that adapts frame rates and resolutions based on content complexity, balancing efficiency and detail. This method enables effective spatial-temporal understanding within a token budget, ensuring comprehensive video representation across varied lengths and complexities.  The pre-training of Seed1.5-VL involved curating 3 trillion high-quality tokens across diverse domains. Image-text pairs from the web were filtered using CLIP scores, size/aspect ratio checks, and deduplication to reduce noise. Using domain-based sampling and duplication strategies, rare visual concepts were overrepresented to address class imbalance. Specialized datasets were added for OCR using annotated and synthetic text-rich images, charts, and tables—object grounding and counting tasks utilized bounding boxes, points, and auto-labeled web data. Additional tasks included 3D spatial understanding using depth annotations, and video understanding through multi-frame captioning, QA, and temporal grounding to support dynamic content analysis.  The evaluation highlights Seed-ViT and Seed1.5-VL’s competitive performance across vision-language tasks. Seed-ViT, despite having significantly fewer parameters, matches or outperforms larger models like InternVL-C and EVA-CLIP on zero-shot image classification tasks, showing high accuracy and robustness on datasets such as ImageNet-A and ObjectNet. Seed1.5-VL demonstrates strong capabilities in multimodal reasoning, general VQA, document understanding, and grounding. It achieves state-of-the-art benchmarks, particularly in complex reasoning, counting, and chart interpretation tasks. The model’s “thinking” mode, which incorporates longer reasoning chains, further enhances performance, indicating its strong ability in detailed visual understanding and task generalization.  In conclusion, Seed1.5-VL is a vision-language foundation model featuring a 532 M-parameter vision encoder and a 20 B-parameter Mixture-of-Experts language model. Despite its compact size, it achieves state-of-the-art results on 38 of 60 public benchmarks and excels in complex reasoning, OCR, diagram interpretation, 3D spatial understanding, and video analysis. It also performs well in agent-driven tasks like GUI control and gameplay, surpassing models like OpenAI CUA and Claude 3.7. The model shows strong generalization to tasks beyond its training scope. The study outlines its architecture, data pipeline, and training methods and identifies future directions, including enhancing tool-use and visual reasoning capabilities.  Check out the Paper and Project Page. 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. The post ByteDance Introduces Seed1.5-VL: A Vision-Language Foundation Model Designed to Advance General-Purpose Multimodal Understanding and Reasoning appeared first on MarkTechPost.

The Famous, Feathered Dinosaur Archaeopteryx Could Fly, Suggests New Study of a ‘Beautifully Preserved’ Fossil The Chicago Archaeopteryx features more soft tissue and delicate skeletal details than any known fossil of its kind, and paleontologists discovered it has a set of feathers key to flight in modern birds The Chicago Archaeopteryx, seen under UV light, shows soft tissues alongside the skeleton. Delaney Drummond Archaeopteryx—a small, feathered dinosaur that lived around 150 million years ago—changed how humans understand the world, “maybe more than any other fossil,” as Jingmai O’Connor, a paleontologist at Chicago’s Field Museum, tells the New York Times’ Asher Elbein. That’s because fossils of Archaeopteryx, which are considered by many to be the oldest known birds, “proved Darwin right” by confirming that all birds are dinosaurs, according to a Field Museum statement. Thanks to an exceptionally preserved and prepared fossil named the Chicago Archaeopteryx recently acquired by the Field Museum, O’Connor and her colleagues have revealed more crucial insight into the species—including that it likely could fly. They detailed their work in a study published Wednesday in the journal Nature. As the Field Museum’s preparators worked on the pigeon-sized Chicago Archaeopteryx, they realized that the fossil included more soft tissues and delicate skeletal details than any other known Archaeopteryx specimen. From previous fossils, they already knew the dinosaur had asymmetric feathers, which are vital to creating thrust in modern, flying birds. But the hard slab of limestone around this specimen had also preserved a key layer of feathers called tertials that had never been documented before in Archaeopteryx. “It’s important that this is the first time these feathers have been seen,” John Nudds, a paleontologist at the University of Manchester in England who didn’t participate in the study, tells the Guardian’s Hannah Devlin. “These new feathers seen in this beautifully preserved specimen—as well as the asymmetric feathers—confirms it could fly.” An illustration shows an Archaeopteryx with tertial feathers that might have helped it to fly. Michael Rothman In modern birds, tertial feathers are crucial to flight, because they cover the space between the bird’s body and its wings. Without them, wings can’t generate lift. In fact, flightless feathered dinosaurs closely related to birds, but not quite birds themselves, didn’t have tertial feathers, according to the statement. Their wing feathers ended at their elbows. While paleontologists don’t know exactly when feathered, landbound dinosaurs gained the ability to fly, most scholars agree that Archaeopteryx is the earliest known bird. “People hypothesized that [the animal] probably had tertials,” O’Connor tells Science News’ Carolyn Gramling. Not only does the Chicago Archaeopteryx finally confirm this hypothesis, but it also indicates that tertials could have played an important evolutionary role in the development of feathered flight, she adds. “Archaeopteryx isn’t the first dinosaur to have feathers, or the first dinosaur to have ‘wings.’ But we think it’s the earliest known dinosaur that was able to use its feathers to fly,” O’Connor says in the statement. That’s despite the fact that “by modern standards, it was a very poor flyer,” she adds to Science News. The Chicago Archaeopteryx is housed at the Field Museum. Delaney Drummond / Field Museum Other notable features the researchers documented include bones in the roof of the animal’s mouth, which provide insight into its skull; scales on the bottom of its toes, which suggest it still spent a lot of time on the ground; and soft tissue on its hands. This last feature indicates that “the third digit could move independently, supporting long abandoned claims from the ’90s that Archaeopteryx could use its hands to climb,” O’Connor explains to Gizmodo’s Isaac Schultz. The team was able to reveal so many telling details thanks to their meticulous fossil preparation, which they conducted with the help of CT scanning and ultraviolet light to make sure they didn’t mistake any part of the fossil for rock and accidentally remove it. A fossil collector discovered the Chicago Archaeopteryx sometime before 1990 in limestone deposits near Solnhofen, Germany, where all known Archaeopteryx fossils originate. It was privately owned until the Field Museum’s purchase. Get the latest stories in your inbox every weekday.

Understand the Full Benefits of a Knowledge Management System In this article:Effective knowledge management is more pressing than ever in today’s dynamic business environment. As organizations strive to capitalize on their collective expertise and data, implementing a robust Knowledge Management System (KMS) can be a game-changer. A KMS not only ensures that valuable information is preserved and accessible but also fosters a culture of learning and innovation. By exploring the full range of benefits a KMS offers, companies can unlock their potential for growth and competitiveness. Below, we delve into how these systems can transform various facets of an organization. 👋 Psst... Did you know you can get unlimited downloads of 59,000+ fonts and millions of other creative assets for just $16.95/mo? Learn more »Unveiling the Comprehensive Advantages of Implementing a Knowledge Management System The adoption of a Knowledge Management System holds the promise of far-reaching advantages for any organization. Primary among these is the ability to capture institutional knowledge — the valuable, often intangible, expertise accumulated over time. A KMS preserves this knowledge, preventing loss due to employee turnover and enabling new team members to come up to speed quickly. Moreover, a KMS serves as a powerful tool for decision-making. With ready access to historical data, case studies, and expert insights, managers and teams can confidently make informed decisions. This repository of organized information ensures that valuable lessons and best practices are recognized and replicated across the organization. Staff development and retention are significantly enhanced by introducing a Knowledge Management System. Employees feel more empowered and engaged when they have the resources to master their roles and contribute meaningfully to the company’s mission. User-friendly KMS platforms facilitate ongoing professional growth, nurturing a more competent workforce. The benefits of a Knowledge Management System extend beyond mere data storage; they include increased innovation. A well-implemented KMS can spark creativity by providing a collaborative environment where ideas are easily shared and refined, leading to more innovative solutions and practices. Enhancing Organizational Efficiency Through Centralized Information Efficiency gains are one of the most tangible outcomes of instituting a Knowledge Management System. A centralized repository of information allows for swift access to needed documents, procedures, and expertise, effectively eliminating the time-consuming process of searching through disparate data silos or querying busy colleagues.Get 300+ Fonts for FREEEnter your email to download our 100% free "Font Lover's Bundle". For commercial & personal use. No royalties. No fees. No attribution. 100% free to use anywhere. An organizational shift to a KMS can markedly streamline workflows and processes. Tasks that once required manual intervention and cross-departmental communication can be automated or simplified, reducing the margin for error and freeing up employees to focus on higher-value activities. This improved operational efficiency often results in cost savings and a better allocation of resources. The consolidation of data within a KMS not only aids in day-to-day operations but also assists in compliance and risk management. By maintaining a clear record of policies, regulations, and compliance documents, businesses can more readily navigate complex legal requirements, mitigating potential risks. Additionally, a KMS can enhance the scalability of operations. As a business grows, its knowledge base can expand seamlessly within the system, providing a robust framework to support the increased complexity and information needs of a larger workforce and client base. Fostering Innovation and Collaboration with Knowledge-Sharing Practices A Knowledge Management System at its core is a facilitator of knowledge sharing, unlocking the collective intelligence of the workforce. When employees freely share insights and expertise, a more informed and agile organization can rapidly adapt to market changes and new opportunities. The cross-pollination of ideas is another benefit that stems from an environment where knowledge flows openly. Employees from different departments or geographies can interact within a KMS, sparking discussions that may lead to breakthrough innovations or more effective problem-solving techniques. Maintaining a competitive edge in today’s economy often depends on a company’s ability to learn and innovate faster than its peers. A KMS creates the infrastructure needed to support continuous improvement initiatives, as lessons learned are systematically captured and made available for future endeavors. Collaboration tools integrated within a KMS can also enhance project management and coordination. Teams can work together on documents, share real-time feedback, and track project progress, ensuring that knowledge is shared and applied effectively. Overall, the strategic implementation of a Knowledge Management System can transform an organization’s approach to information sharing, decision-making, and innovation. It improves efficiency, drives performance, and establishes a foundation for continuous learning and competitiveness in the ever-evolving business landscape. Jack Nolan Jack Nolan is a freelance graphic designer with over 10 years of experience helping brands stand out through bold, impactful design. Specializing in logo design, visual identity, and digital illustrations, Jack has worked with startups, small businesses, and global clients to bring creative ideas to life. His passion for clean, timeless design is matched only by his commitment to understanding client needs and delivering work that exceeds expectations. When he's not designing, Jack enjoys hiking, experimenting with photography, and exploring the latest trends in design. 10 Must-Have Travel Tools for Designers in 2025Traveling as a designer in goes beyond simply stuffing your passport and sketchbook. Modern designers must remain connected, motivated, and...5 Steps to Respond to an Interview Request EmailCongratulations! You’ve successfully navigated the application process and landed an interview request. It’s an exciting milestone in your job search...

In this guest post, Owlchemy Labs Accessibility Product Manager II Jazmin Cano dives deep into how the team used Unity to develop an all-new vision accessibility update for Cosmonious High in virtual reality (VR).At Owlchemy Labs, we have made a dedication in our Accessibility Statement to the pursuit of creating VR games for everyone. Historically, our team has prioritized making significant strides in unsolved areas of accessibility in VR including subtitles, physical accessibility, and height accessibility. We encourage every developer to create with accessibility in mind, and actively incorporate accessibility into our games at launch and through post-launch updates.For our last accessibility update to Cosmonious High, we added a range of gameplay options, including one-handed controller mode, seated player mode, subtitling, and more. If a game can only be played by an audience with no disabilities, it is a sign of an unfinished product. We take pride in forging new paths through research, testing, and partnerships to ensure gaming is available to the widest possible audience.When we started our research in developing for blind and low-vision players, we found that the main issue was the inability to follow a storyline or in-game directions.Oftentimes, there is no subtext available for all of the scenes in a game, and different scenes are used to advance the story. A storyline could be complex and challenging to follow without context or subtext. Other aspects of a game, like completing a puzzle or a task, become difficult when not all pieces are clearly visible for players or when vital clues are given without any text or visual queues.From our research, we found it was important to create a way for players to receive game information through audio descriptions using Text To Speech (TTS). Most people who are blind or have low vision use assistive technology called screen readers, which allows them to navigate screens with a keyboard, and their computer or phone will read out the text. Our game features are similar in the sense that they allow players to navigate virtual spaces using their hands or a joystick as methods for reading out objects and descriptions. Each feature is built to follow expectations regarding screen-reader standards, such as ducking non-TTS audio and allowing the user to cancel TTS mid-description.We haven’t seen narration or TTS in other released VR games before, so we knew launching our approach for accessibility in this area – after researching best practices and playtesting with those who are legally blind or have low vision – would benefit the industry.Developing for blind and low-vision players isn’t just adding audio cues and options. Blindness is a spectrum, and creating gameplay in a way that is fully inclusive of blind and low-vision players also includes feeling and seeing.To accomplish this, we developed more haptic feedback as a method to identify an object when a player highlights it. This is particularly important when a player is completing a puzzle. Players can identify an object and feel it when they have selected it.We’ve also added high-contrast object highlighting, which outlines key objects in the environment, making it easier for players to see the object and understand their selection with haptic feedback.To bring this all together, we incorporated a Grab-and-Release confirmation. Using this, players will receive haptic feedback when highlighting an object that is shown through high-contrast object highlighting, and audio will play to inform them when an object has been grabbed and released.When we started developing this update, we anticipated a few challenges, ranging from design decisions to playtesting. Thankfully, we had a great start to the project after meeting with Steve Saylor, a video game accessibility consultant who is blind. Consulting with Steve, we were able to identify the features we would need and what expectations someone with low vision would have. We did a lot of research, experimentation, and testing to determine what worked best and what we could execute successfully.For the vision accessibility update, we knew we needed to branch out when finding playtesters with low vision to help provide valuable feedback and help with our design. We teamed up with VROxygen to find a group of playtesters with blindness and low vision to provide feedback on our iterations and help us prioritize improvements we needed to make, which worked out well. Opening testing to remote players from anywhere in the world allowed us to get a wide range of perspectives on the project, with feedback coming from people with varying levels of vision.The path to creating this update offered some challenges to consider. Sometimes, what seems like the most obvious answer is not always the best one, especially when developing for accessibility. We took the time to work through each aspect of this update to make sure the features we were adding worked well for those who needed it most.When we enabled TTS early in the project, we started with automatic narration. This meant that any object a player’s hands waved over would be described, even if that meant speaking on top of a previous description that may still be going. For audio descriptions to be valuable, they need to be heard without other audio fighting for priority.This resulted in a few changes that worked well in playtesting. For example, we decided to add a button press to activate descriptions instead of having them read automatically, which led to a more comfortable experience. This gives players agency when deciding if a description is read to them when their hand is placed over an object or pointing at something in the distance. It also prevents accidental TTS from happening if a player moves their hand over objects they didn’t mean to have read.While TTS describes the name of the object and a visual description, it doesn’t take very long. Even with short descriptions, though, we know people would want behavior similar to a screen reader (i.e., the ability to cancel audio while it’s being read).Another thing we learned with regards to audio is the importance of lowering it so that the lower volume allows the TTS audio descriptions to have priority. We ran into a problem where TTS can be triggered during an interaction with an NPC, making the NPC’s dialog quieter and easy to miss. At this time, players are not able to “rewind” or “retrigger” the same audio to play; however, players can wave their hands and NPCs will respond back to help.But one of the hardest parts about building features like these is making sure they will actually help users who need them. The best way of determining usefulness for new accessibility features is through testing. Of course, being able to quickly make new builds for all of our platforms after each round of feedback and development was essential to making this update the best it could be. One unlikely tool we found useful for fast iteration on our designs was Unity's Post-Processing Stack.Before sending builds to playtesters, our developers wanted to test the effectiveness of features internally. Since many of our developers are sighted, we used the Post-Processing Stack and created entries in our debug menu that allowed us modify the visual clarity of what we were seeing in the headset. This helped our developers simulate roughly what it is like to have different levels of reduced vision while playing the game. Since we could now rapidly identify and tackle the most obvious issues, we were able to iterate on designs more quickly and make sure we were getting the most out of the external playtest sessions with blind and low-vision testers.In developing this update, we learned good design practices for blind and low-vision accessibility that we’ll use in future games. One of those learnings comes from an approach that started long before this update.Our developers create with accessibility in mind, and we’ve learned from each accessibility update that starting a project with this approach makes developing future accessible features much easier. For example, making sure objects have proper names and descriptions that are useful as alt text on images requires less effort and saves time.Having large objects and large text with good contrast makes a lot of a VR game’s world easier to see and read. Many players who described their vision as highly blurry could lean in and read much of the game’s text. Players who couldn’t fully make out text would then use the assist button to hear the words as needed.The text in Cosmonious High is stationary, so players don’t need to worry about it being at a fixed distance from their face, moving away from them as they continue to lean forward. One player commented that we made the game accessible before even thinking about vision accessibility as he described the size and colors in the world not specific to this update. While we appreciate the compliment, the design practice of making things larger and easier to distinguish is part of our development process – it’s in our developer documentation.Specific to this update, allowing players to decide when to get audio descriptions, allowing players to cancel audio, and keeping descriptions short with valuable information is key to giving players agency.Developing for blind and low-vision players has added to our arsenal of development tools. We hope to build upon our learnings in all areas of accessibility and plan to launch future titles with more accessibility included. We’re excited to share this update of Cosmonious High as a continuation of our mission to make VR for everyone.Owlchemy Labs’ Cosmonious High is available on multiple platforms. Check out more blogs from Made with Unity developers here.

Researchers are trying to determine the identity of the shipwreck. Image: Screenshot, CBC/YouTube Get the Popular Science daily newsletter💡 Breakthroughs, discoveries, and DIY tips sent every weekday. A Canadian kid is proof that major scientific discoveries don’t always have to come from grizzled researchers with fancy equipment. Two years ago, then-8-year-old Lucas Atchison went on a family trip to Point Farms Provincial Park in Ontario. Armed with a metal detector he had just received as a birthday present, Atchison dutifully scanned the area, hoping to hear that coveted “beep.”  Eventually, he did. Eagerly digging into the site, Lucas uncovered a metal spike, which his father initially dismissed as something used to tie up boats. But the budding archaeologist insisted they dig further. Soon, they uncovered another spike attached to a piece of wood. The father-son pair had likely stumbled on a two-century-old shipwreck.  “We were on the beach, we got our metal detector out, and as soon as we set it up, ding! It was a spike from the shipwreck,” Lucas recently told the CBC. Lucas’ find led to a full-on excavation  Once the duo realized what they had found, they alerted the provincial parks staff, who then contacted the nonprofit Ontario Marine Heritage Committee (OMHC). The OMHC met with Lucas in 2023, but it took several years to obtain the necessary regulatory permits to excavate the area further. Finally, earlier this month, they began digging. Though still in the early stages, marine archaeologist Scarlett Janusas told the CBC that the shipwreck’s “double frames” suggest it is of the schooner variety. Work still needs to be done to definitively confirm the ship’s identity, but archaeologists working on the project believe they may have a likely contender in the St. Anthony. That vessel wrecked near Lake Huron in 1856 during a trip transporting wheat between Chicago and Buffalo. Researchers at the site are currently creating sketches of the wreck from above and from the side to analyze it in more detail. They are also reportedly reviewing 19th-century insurance requirements for ships, which could provide clues about the vessel’s identity and when it sank. Those documents outline specifications for ships at the time, such as the number of fasteners required in the frames. By cross-referencing those details against records of missing ships, they hope to definitively identify the wreck. You never know what you might find  Lucas’ discovery is a nice reminder that, even today, in our seemingly mapped-out, modernized world, you never know what you might find with a trusty metal detector. His case is far from unique. In just the past few months, a Romanian man used a metal detector to uncover an ancient hoard of Roman coins. Even more recently, a Scottish man discovered a 4,000-year-old axe buried on his family farm. Sometimes, people may not even realize the significance of their finds until much later. Case in point: an Australian prospector named David Hole found a large rock with a metal detector in 2016, believing it to be a gold nugget. Only recently did he learn that it was actually a far rarer 4.6 billion-year-old meteorite. As for the Ontario shipwreck, researchers say they eventually plan to re-bury the find in an anaerobic, oxygen-free environment. That setting could help preserve it by protecting it from natural degradation caused by parasites and other organisms. Lucas, meanwhile, may be just beginning a long journey with his trusted metal detector.

LAI #75: Generative AI vs. Agentic AI vs. AI Agents 0 like May 15, 2025 Share this post Author(s): Towards AI Editorial Team Originally published on Towards AI. Good morning, AI enthusiasts, This week’s issue dives into where the field is heading — beyond generation, toward autonomy and better error awareness. We’re starting with a breakdown of the increasingly fuzzy but important distinctions between Generative AI, Agentic AI, and AI Agents. Then we move into applied innovation: Microsoft’s GraphRAG, multimodal RAG systems using Cohere and Gemini, and a practical framework for predicting when your model is about to get something wrong. Also in the mix: DNNs vs. tree-based models for e-commerce ranking, a powerful Cursor.ai-like browser extension from the community, and this week’s poll on when vibes are enough — and when accuracy has to come first. Let’s get into it. — Louis-François Bouchard, Towards AI Co-founder & Head of Community Learn AI Together Community Section! Featured Community post from the Discord Retconned has built Sophon, an AI chat app that enhances your browsing experience by understanding and interacting with your tabs. With its intelligent composer, it can see the tabs you have open, allowing it to understand context and autofill forms, textboxes, or fields with a single click. It is a browser extension and completely free. Check it out here. Share your feedback in the thread and support a fellow community member! AI poll of the week! Most of you are doing vibe checks, and of course, for general tasks, the entire idea is for the AI to not feel like AI. But would you also rely on “vibes” for more quantitative tasks, where output accuracy matters more than output feel? Share in the thread, let’s decide together! Meme of the week! Meme shared by rucha8062 TAI Curated Section Article of the week How GraphRAG Works Step-by-Step By Mariana Avelino This blog explains Microsoft’s GraphRAG, a method that uses knowledge graphs for retrieval-augmented generation. Key detailed processes were graph creation, involving entity extraction, community partitioning, and querying, with distinct Local and Global Search functions. It outlined how entities, relationships, and community reports are generated and used for LLM response generation, including context management and semantic retrieval. Our must-read articles 1. Distill-then-Detect: A Practical Framework for Error-Aware Machine Learning By Shenggang Li The author presented a framework, “Distill-then-Detect,” to address prediction errors in machine learning models, particularly the “big misses” on critical data slices. This approach involves distilling a compact “student” model from a larger “teacher” model. It then quantifies teacher uncertainty and trains a meta-model to predict where the teacher is likely to err. By combining these signals into a risk score and applying conformal calibration for thresholding, the system effectively flags high-risk predictions. Experiments demonstrated that this method identified error-prone cases with balanced precision and recall while clustering these errors provided actionable insights into problematic data segments. 2. Beyond Text: Building Multimodal RAG Systems with Cohere and Gemini By Sridhar Sampath Traditional Retrieval-Augmented Generation (RAG) systems often fail to process visual data. This article details a multimodal RAG system designed to overcome this limitation by understanding both text and images within documents. It utilizes Cohere’s multimodal embeddings to create unified vector representations from content like PDFs. Gemini 2.5 Flash then generates context-aware answers using either matched text or images, with FAISS managing vector indexing. It explains the system’s workflow, from document upload to answer generation, demonstrating its enhanced capability to extract information from charts, tables, and other visuals compared to text-only RAG. 3. Generative AI vs. Agentic AI vs. AI Agents: What Everyone Needs to Know By Poojan Vig The article clarified the distinct roles of Generative AI, Agentic AI, and AI Agents. It explained that Generative AI produces new content based on learned patterns. Agentic AI focuses on strategy, planning, and iteration towards a goal without continuous human intervention. AI Agents then sense their environment and execute actions in the digital or real world. Using a cooking analogy and examples like automated customer service, the piece illustrated how these AI types can operate independently or collaboratively to perform complex tasks. 4. DNNs vs Traditional Tree-Based Models for E-Commerce Ranking By Nikhilesh Pandey The author discusses the evolution of e-commerce ad ranking systems, detailing the shift from traditional tree-based models to Deep Neural Networks (DNNs). It outlines why tree-based models have reached their limits and how DNNs offer superior capabilities for handling complex data, personalization, and achieving better conversion rate (CVR) predictions. Using DoorDash Ads as a case study, the piece illustrates the iterative migration process, including defining baselines, optimizing model training and evaluation with techniques like data normalization and distributed processing, and addressing challenges such as the offline-online performance gap. If you are interested in publishing with Towards AI, check our guidelines and sign up. We will publish your work to our network if it meets our editorial policies and standards. 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 Towards AI - Medium Share this post