Joshua To

🚀 Big news for AI agent builders: Agent Reinforcement Trainer (ART) now integrates seamlessly with 🦜🔗 LangGraph! If you’ve ever built multi-step, ReAct-style agents, you know the pain: endless prompt tweaking, fragile reasoning, and lots of manual trial-and-error. With this new integration, you can now train your LangGraph agents with reinforcement learning — making them smarter, more adaptive, and more efficient over time. ✨ What this means in practice: * Agents learn to reason through multi-step problems without hand-holding * They improve tool usage automatically (no more prompt spaghetti) * They make smarter decisions in complex workflows * Training is scalable and robust, powered by ART’s reinforcement engine And the best part? It’s a drop-in upgrade: * Swap in ART’s init_chat_model and wrap_rollout * All interactions, tool calls, and reasoning chains get logged automatically * Plug straight into ART’s RULER reward function for training — no custom rewards needed 💡 Imagine your LangGraph prototypes evolving into production-ready agents that actually get better with use. That’s the power of combining LangGraph’s agent framework with ART’s reinforcement learning. 👉 Full docs & examples here: https://lnkd.in/gwWj-xxx Can’t wait to see what the community builds with this. If you’re experimenting with multi-step AI agents, give your agents the ability to learn from experience!

1,890

81 Comments

NEW 🤘 Owkin MEGA COOL MODEL RELEASE. Open for the community of course We’re thrilled to publicly launch HistoPLUS, a state-of-the-art model for pan-cancer nuclei detection, segmentation and classification in H&E-stained pathology images. The model weights and an open implementation are now available. Key Highlights: 🔬 Unlocks robust detection, segmentation & classification of 13 cell types — including understudied immune populations such as neutrophils and eosinophils. 📈 Outperforms SOTA with 5% better detection and 24% higher F1 classification, using 5x fewer parameters. 🌍 Generalizes across multiple cancer types. 🧑🏼‍💻 GitHub Repository: https://lnkd.in/eGFCa-c9 🤗 Model Weights: https://lnkd.in/eFpBB7x4 📄 Preprint: https://lnkd.in/etPtSKph HistoPLUS is built on top of the powerful H0-mini foundation model, developed in collaboration with our dear Bioptimus’ team. This study also makes use of data generated by the MOSAIC research consortium Gustave Roussy, CHUV / Centre hospitalier universitaire vaudois, Universitätsklinikum Erlangen, Charité - Universitätsmedizin Berlin, University of Pittsburgh 🙏 Huge thanks to the expert pathologists whose invaluable annotations made this work possible—

438

13 Comments

Today, we announced the open sourcing of our AI tools for Online safety at Roblox. We invest significantly in this area in building the tools and running it on our platform spanning Core, Edge, and Cloud to enhance safety for our player community. The need for improving Digital safety in the world of AI is substantial for everyone, and this initiative marks a significant step in that direction that everyone can benefit from. #AISafety #Roblox #DigitalSafety https://lnkd.in/g_bftZCb

59

3 Comments

"User interfaces are gonna go away." : Eric Schmidt He said it out loud on the Moonshot podcast by Peter Diamandis, but it’s something we’ve believed for a long time at Thesys. Traditional interfaces are dumb and rigid. They were built for linear, pre-determined user journeys. They were not built for systems that think, learn, or respond dynamically. Over the past two years, we’ve been heads down building the first practical implementation of Generative UI. Today, our customers use it with the simplicity of integrating an API. Also, Generative UI isn’t just about better user experience and engagement. It also drastically improves execution efficiency of teams as they - ✦ Skip manually building the frontend, saving months ✦ Release faster and reduce time to market ✦ Maintain and scale the UI easily with every new feature or flow Interested in learning more about Generative UI? Dropping a few resources in comments.

50

3 Comments

We’ve released our research vision. The key insight (which has been the through line for poolside since our founding): Reinforcement Learning is the most important scaling axis to AGI In a world full of noise about AI's future, we have a specific view on how we’ll get to AGI through software. We’re on the path Read our full research vision here: https://lnkd.in/gmzYrT6S

257

12 Comments

Elie Schoppik has spent his career at the intersection of education and engineering. He built early ed-tech tools that helped people actually retain knowledge, founded a Bay Area coding bootcamp that trained thousands of developers through practice and feedback, and now leads Technical Education at Anthropic. Despite all that experience, Elie says most teams are using AI wrong: they’re treating it like an all-knowing oracle instead of a tutor. This is why Elie designs prompts, examples, and workflows that treat AI as a teacher, instead of a shortcut. In our conversation, we discuss: • Why relying on AI for “fast answers” erodes judgment and verification • How to turn prompts into guided lessons • Why context engineering is the most underrated skill in 2025 • Practical exercises any team can run tomorrow to raise AI fluency • What leaders should steal from Anthropic’s approach to technical education  Episode is live now — full conversation linked in the comments.

30

2 Comments

At last year’s NEXT Conference, I had the privilege of speaking with Ivan Poupyrev, CEO & Founder of Archetype AI—a man on a mission to make AI not just think, but sense and interact with the world around us. With Newton, Archetype AI is building a foundational model that interprets real-world sensor data—radar, cameras, vibrations, even temperature. This isn’t just AI for digital interactions; it’s AI that truly understands the physical world, unlocking new possibilities across industries. Ivan’s journey from Google’s Project Soli (where he pioneered micro-gesture control) to Archetype AI reflects a broader shift: Instead of humans adapting to technology, technology is finally adapting to us. And with AI evolving at breakneck speed, this shift might happen sooner than we think. What excites me most? Physical AI is set to redefine industries—from automotive and consumer electronics to smart environments and beyond. Companies that embrace it early will shape the future of human-machine interaction. A huge thanks to Ivan for this inspiring conversation—and for indulging my “world’s smallest violin” moment! 🎻 If you’re curious about where AI meets the real world, this is one interview you don’t want to miss. Watch the full discussion on YouTube (link in comments). #AI #PhysicalAI #ArchetypeAI #NEXTConference #Innovation #HumanMachineInteraction #AccentureSong

55

4 Comments

While we haven't officially launched, can't help but be excited to see this great artcle in the @WSJ featuring LGND AI, Inc.(http://lgnd.io) and the opportunity for AI applied to earth observation data for all sorts of important uses. This is a huge and important data source that is vastly under represented in the current AI platform shift - and we aim to change that. Big stuff ahead! https://lnkd.in/ghd76mBB

89

9 Comments

At CES 2025, Jensen Huang made a prediction that's reshaping how we think about enterprise software: "The IT department of every company is going to be the HR department of AI agents in the future." This statement isn't just provocative—it's profoundly insightful about where software development is headed. Think about what HR does: recruiting talent, onboarding employees, managing performance, and optimizing teams. Now imagine IT professionals doing the same with AI: - Instead of writing code from scratch, they'll recruit pre-built AI capabilities. - Rather than configuring every setting, they'll onboard AI agents through natural language. - Instead of debugging endlessly, they'll manage the performance of AI systems. - Rather than working in isolation, they'll orchestrate teams of AI and human contributors. This shift is why AI-powered visual development platforms are rapidly replacing traditional low-code tools. They represent the perfect implementation of Huang's vision—letting IT professionals truly operate as strategic "HR managers" for their digital workforce. The most powerful platforms combine three essential elements: 1. Natural language interfaces that let you create by simply describing what you need. 2. Visual refinement tools that provide intuitive control over AI-generated elements. 3. Flexibility to seamlessly move between AI assistance, no-code building, and code customization when needed. Legacy low-code platforms built without AI at their core are struggling to adapt, while solutions that are AI-only often lack the enterprise-grade security and visual customization capabilities that organizations require. As internal tools increasingly move toward this AI-powered future, how is your organization preparing IT teams to become effective "HR managers" for your growing AI workforce? #ai #lowcode #nocode #opensource

73

1 Comment

Headlines abound about the 95% failure rate of ‘AI projects.' And while it’d be fun to pile on against exaggerated hype, the real failure here is our collective blindspot in understanding just how low the odds of success are for ANY custom-built enterprise project. Discussed are the reasons why enterprise ‘AI’ projects are particularly fraught, but not far worse than technology builds over the last 30 years: https://lnkd.in/eFD6kbrA

230

13 Comments

Every breakthrough comes with hidden limitations. Right now, embeddings—the foundation of modern RAG systems—stands close to a ceiling. While embeddings are excellent at representing data in vectors they fall short in capturing rich, interconnected relationships—the kind of nuanced, human-like understanding we need for next-gen AI. Knowledge graphs solve this. They preserve information the way humans process it—as interconnected concepts rather than isolated data points. They’re perfect for tackling what embeddings miss: multi-hop reasoning, hierarchical structures and temporal logic. 2025 could be the year knowledge graphs go mainstream, reshaping how AI systems learn and reason. If you’re in AI or data, this is your signal: start exploring knowledge graphs now. What are you betting on in 2025?

29

10 Comments