Dawn Song

I had the great honor and pleasure of sitting down with @JeffDean for his first public talk since leaving Google, where he spent an extraordinary 27 years. Few people have shaped modern computing and AI as profoundly - from MapReduce and Bigtable to TensorFlow, Mixture-of-Experts, TPUs, and Gemini. Our conversation covered some of the biggest questions shaping the future of AI: • How do you recognize a foundational idea before everyone else does? • How do you choose a research problem worth spending 5 years on? • What can coding teach us about building better reasoning models? • What might recursive self-improvement (RSI) actually look like? • What happens when the scientific discovery loop itself becomes increasingly automated? (and how is Jeff’s new startup going to contribute in this space?) • As AI becomes increasingly autonomous, how do we keep it safe and secure? • What should the next generation of researchers be working on? Here are some key insights and highlights for anyone building the future of AI. 🧵1/8

2/8 — What might recursive self-improvement actually look like - and what is Jeff building next with Discovery Loop? Jeff’s view: using ML to improve ML isn’t new. Neural Architecture Search and the Evolved Transformer were early examples. What changes now is the scope. Imagine increasingly automating everything that goes into improving AI - architecture, data, evals, experiments, feedback - so more and more of AI R&D itself becomes automated. Jeff pointed out that the same basic loop applies much more broadly to science and engineering: break down the problem → propose an approach → experiment → evaluate → learn → repeat Now imagine AI automating more of this loop — running thousands of experiments in parallel, compressing iterations from days or weeks to minutes or hours, and using the results to choose better experiments next. That’s the vision behind Jeff’s new startup, Discovery Loop: automate machine learning, science, and engineering to dramatically accelerate discovery. The bigger idea: use AI not only to improve AI, but to increase the rate at which we discover new things. 🔬

Vídeo

Introducing CUA-Lite 🧵 — an open platform for computer-use agents. Training and benchmarking CUAs (Computer-Use Agents) requires four core pieces: 1️⃣ Agents — the models and the scaffolding that drives them 2️⃣ Environments — runtime/sandboxes for agents to interact with, tasks & verifiers/graders 3️⃣ Traces — records of agent trajectories 4️⃣ Frameworks — to evaluate, SFT & RL-train agents Today, all four are fragmented. Every agent ships with its own implementation, often in a separate repo — there is no unified way to run them all. Every environment exposes its own interface and action space, often requiring an expensive VM sandbox for each verifiable task. Traces come in incompatible formats. And without common standards across the stack, every project ends up rebuilding its own tooling/framework for eval, SFT, and RL. CUA-Lite unifies the stack: → One standardized interface & action space for agents and environments → One standardized format for agent traces → One framework for evaluation, SFT & RL → Across desktop, browser & mobile And open resources plug straight in, creating the largest open collection of CUA agents, environments and traces, all in a unified format: 🤖 10+ CUAs, including GPT, Claude, Gemini, Qwen, Muse-Glimmer, UI-TARS 🌐 15+ benchmarks, including OSWorld, WebArena & AndroidWorld ⚡ Optional VM-free sandboxes with 30K+ verifiable tasks for training 📚 10+ trace datasets, freely available on Hugging Face, including public datasets converted into the standardized format and fresh rollouts from frontier open-weight CUAs Led by @BerkeleyRDI , our goal is for CUA-Lite to become a community-driven, open-source ecosystem for computer-use agents. Join the community and contribute today: bring an environment (runtime/sandbox + tasks + verifier), traces, or an agent, and plug it into CUA-Lite!

GIF

[2/10] Learn more about CUA-Lite, an open platform for computer-use agents, and join the community: 👇 Site: t.co/N6uMHTfNFr Code: t.co/NtA9x1xs3k Data: t.co/yRybcZe2PN Leaderboard: t.co/ANWTWO6l3L Huge thanks to my PhD student Zhanhui Zhou @asapzzhou for spearheading the project!

I completely agree. This may be one of the clearest signs yet that we’re entering a new era of AI agents. We developed ExploitGym to measure whether AI agents can turn real-world vulnerabilities into working exploits. But what happened during the evaluation at OpenAI went beyond what the benchmark was designed to measure: agents found unintended paths, coordinated across instances, worked around containment controls, and compromised real-world infrastructure. As agents become more capable and autonomous, ensuring they are properly aligned, that they understand and respect the boundaries of what they are authorized to do, is becoming increasingly critical. Lots to do as next step: we need to continue measuring frontier cyber capabilities as they evolve, and urgently strengthen agent alignment, monitoring, containment, and secure evaluation/training infrastructure. These are becoming essential safeguards as agent capabilities rapidly advance. The question is no longer just what AI agents can do. It’s what they will do when we haven’t anticipated the path they take.

More info on ExploitGym and the OpenAI/Hugging Face incident: openai.com/index/hugging-… cybergym.io

🤔 Wow, this looks quite unexpected: ❤️ 343 · 🔁 78 · 🔖 257, with only 30K views! Either X is remarkably good at finding exactly the right audience, or it’s failing to recognize that this post deserves broader reach. Which is it? Poll below 👇

What’s the explanation?

1/ Can AI agents build formally verified software repositories? Introducing Vero: the first benchmark for joint implementation and proof synthesis at the repository level. As AI agents write a growing share of our software, we need more than just working code; we need machine-checked guarantees of correctness and security. In Vero, we find that repository-scale verified code generation is still out of reach: the strongest frontier agent fully verifies only 27 of 43 real-world repositories. Website: t.co/LhDcK3gnLT

2/ The Challenge: AI agents are impressive at writing code, but can we trust it at the scale of real software? Existing benchmarks either target single functions or evaluate proof generation on a fixed implementation. Real verified software (OS kernels, cryptographic protocols, distributed systems) lives in multi-module repositories, where code, specs, and proofs are deeply interdependent and a single refactor can invalidate proofs across the codebase. Yet no benchmark today evaluates joint code-and-proof generation and formal verification at this scale.

Huge congratulations to @Zai_org on GLM-5.3! Thrilled to see GLM-5.3 pushing the open-source frontier on both CyberGym and ExploitGym. We built these benchmarks to measure frontier cyber capability, track capability transitions as models improve, and provide evidence to inform responsible deployment. As our work has shown, frontier cyber capabilities are advancing rapidly. It's critical that our evaluations, safeguards, and governance evolve just as quickly. We discussed these challenges in-depth at our recent Agentic AI Summit. Check out the session recordings and join the conversation: t.co/UJlZNBE4Wp.

🚀 The future of AI is agentic — this is one message echoed across every stage at the Agentic AI Summit 2026 (Aug 1 & 2), the largest gathering dedicated to agentic AI: 🏛️ ~5,000 attendees in person at UC Berkeley 🌍 ~100,000 joined online from around the world 🎤 ~200 world-class speakers plus ~200 poster presentations, from frontier AI researchers and visionary founders to leaders and pioneers across academia and industry. 💡 Here are just a few glimpses into the ideas that shaped the conversations at the summit: 💬 "The text box is AI's radio-on-TV phase…. Every new medium starts by imitating the old one." — Peter Steinberger @steipete, Creator of OpenClaw, OpenAI 💬 "AI infrastructure isn't a chip problem. It isn't a model problem. It's a systems problem." — Peter DeSantis, SVP, Foundational AI Models, Custom Silicon, Quantum Computing, Amazon 💬 "There will be no AI job apocalypse... we just can't find enough skilled AI engineers." — Andrew Ng @AndrewYNg, Founder, DeepLearning .AI 💬 "Coding capabilities and cyber capabilities are two sides of the same coin—you cannot make models better at coding without also making them better at cyber." — Dawn Song, Professor, UC Berkeley; Co-Director, Berkeley RDI; VP of AI Research, Meta Superintelligence Labs 💬 "'Curfew' comes from the French word for extinguishing fire. Medieval cities tried to restrict fire, yet London still burned. AI resilience won't come from one breakthrough. There's no silver bullet for AI safety—only an ecosystem." — Wojciech Zaremba, Co-Founder, OpenAI 💬 "This is the biggest scientific bet our civilization has ever made—bigger than the Apollo program, the internet buildout, and the Manhattan Project combined." — Jasjeet Sekhon, Chief Strategy Officer, Google DeepMind 💬 "Our generation was too late to explore the Earth, too early to explore the stars—but right on time to build superintelligence." — Richard Socher, Founder/CEO, Recursive Superintelligence 💬 "I genuinely believe the next two years will be the time of architecture—the biggest gains will come from stepping away from transformers." — Jerry Tworek, CEO, Core Automation; Former VP of Research at OpenAI 💬 "Recursive Self-Improvement isn't one capability. It's four: Ideation, Implementation, Experimentation and Evaluation.." — Oriol Vinyals @OriolVinyalsML, Former VP of Research, Google DeepMind; Co-Founder, Discovery Loop 💬 "We are in a capability overhang—models are far more capable than they're able to side-effect into the world today." — Ryan Lopopolo, Principal Engineer, Agentic Google Cloud Platform; Previously Led Dark Factory at OpenAI 💬 "Stop thinking about evaluation as the last check before shipping—think of it as an engine that helps you ship a better agent every single day." — Michele Catasta, President and Head of AI, Replit 💬 "The bottleneck becomes your attention as an agent-using engineer. … We're moving from seeing the code to seeing the entire business." — Alex Graveley, Co-Founder of FlyingObject .ai; Co-creator, GitHub Copilot & Perplexity Computer 💬 "An agent isn't just an LLM — it's an LLM surrounded by what I call infrastructure... another word for that is computer science." — Jonathan Cohen, VP of Applied Research, Nvidia; Academy Scientific and Technical Award Winner 💬 "We don't arbitrate the truth. We give people the most powerful tools to make up their own minds." — Chris Bregler, Senior Director / Distinguished Scientist, Google DeepMind; Academy Scientific and Technical Award Winner 💬 “Thinking doesn’t have to be in text! ... We can even use multiple modalities simultaneously to “think” at the right level of abstraction for the problem at hand” — Sergey Levine, Co-Founder, Physical Intelligence; Professor, UC Berkeley 💬 “Video is the most general modality that we have that allows us to simulate real-world experience." — Anastasis Germanidis, Co-Founder/Co-CEO, Runway 💬 “The relationship between AI and enterprise data is not one-directional. Understanding both sides of that equation is the difference between AI that works and AI that disappoints.” — Dan Roth, Chief AI Scientist, Oracle; Professor, UPenn 💬 "Maybe 99.9% of training data in the next step will be synthetic." — Weizhu Chen, Technical Fellow & CVP, Microsoft AI 💬 "It's maybe the best time ever to start a company—but most 'obvious' AI products will be outcompeted by the frontier labs. The real opportunities lie in solving specific customer problems." — Alfred Lin, General Partner, Sequoia Capital ✨Over two days, we explored one central question: How do we build AI systems that are not only more capable, but also more trustworthy, more secure, and ultimately more beneficial for humanity? This wasn't the end of a conference - it was the beginning of the next chapter for agentic AI. Join us to shape and steward the future of AI for human flourishing! 🙏 A heartfelt thank you to our speakers, sponsors, volunteers, partners, and every attendee (in-person or online) who made this summit possible. 👇 What are your learnings, insights, favourite talk, quote, or moment from the summit? We'd love to hear it below!

🎥 Missed the summit? Watch session recordings online: t.co/pV9eHyxZrt We’re deeply grateful to all of our sponsors and partners for supporting this community and helping bring the Summit to life:@daytonaio, @ema_unlimited, Fetcherr, @LambdaAPI, @replit, @amd, @circle, @databricks, @digitalocean, Novara, @getpostman, @pytorch, @snorkelai, @SUSE, @temporalio, @alice_dot_io, @lobehub, @LudoRobotics, @McKinsey, @nutanix, @OutSystems, @SamsungNext,@vesslai, @trustvijil, @across_ai, @elastic, @GoKiteAI, @neo4j, @novita_labs, @fastinoAI, Valency, @GoogleDeepMind, @lightspeedvp, @MayfieldFund, @NEA.

1/2 🚀 The future of AI is agentic — this is one message echoed across every stage at our Agentic AI Summit 2026 (Aug 1 & 2), the largest gathering dedicated to agentic AI: 🏛️ ~5,000 in-person attendees at UC Berkeley 🌍 ~100,000 joined online from around the world 💡 Just a few glimpses into the ideas that shaped the conversations at the summit: 💬 "The text box is AI's radio-on-TV phase…. Every new medium starts by imitating the old one." — Peter Steinberger @steipete, Creator of OpenClaw, OpenAI 💬"AI infrastructure isn't a chip problem. It isn't a model problem. It's a systems problem." — Peter DeSantis, SVP, Foundational AI Models, Custom Silicon, Quantum Computing, Amazon 💬 "There will be no AI job apocalypse... we just can't find enough skilled AI engineers." — Andrew Ng @AndrewYNg Founder, DeepLearning . ai 💬"'Curfew' comes from the French word for extinguishing fire. Medieval cities tried to restrict fire, yet London still burned. AI resilience won't come from one breakthrough. There's no silver bullet for AI safety—only an ecosystem." — Wojciech Zaremba Co-Founder, OpenAI 💬 "This is the biggest scientific bet our civilization has ever made—bigger than the Apollo program, the internet buildout, and the Manhattan Project combined." — Jasjeet Sekhon, Chief Strategy Officer, Google DeepMind 💬 "Our generation was too late to explore the Earth, too early to explore the stars—but right on time to build superintelligence." — Richard Socher @RichardSocher, Founder/CEO, Recursive Superintelligence 💬 "I genuinely believe the next two years will be the time of architecture—the biggest gains will come from stepping away from transformers." — Jerry Tworek, CEO, Core Automation; Former VP of Research at OpenAI 💬 "Recursive Self-Improvement isn't one capability. It's four: Ideation, Implementation, Experimentation and Evaluation.." — Oriol Vinyals, Former VP of Research, Google DeepMind; Co-Founder, Discovery Loop 💬 "We are in a capability overhang—models are far more capable than they're able to side-effect into the world today." — Ryan Lopopolo, Principal Engineer, Agentic Google Cloud Platform; Previously Led Dark Factory at OpenAI 💬 "Stop thinking about evaluation as the last check before shipping—think of it as an engine that helps you ship a better agent every single day." — Michele Catasta, President and Head of AI, Replit 💬 "The bottleneck becomes your attention as an agent-using engineer. … We're moving from seeing the code to seeing the entire business." — Alex Graveley, Co-Founder of FlyingObject .ai; Co-creator, GitHub Copilot & Perplexity Computer 💬 "An agent isn't just an LLM — it's an LLM surrounded by what I call infrastructure... another word for that is computer science." — Jonathan Cohen, VP of Applied Research, Nvidia; Academy Scientific and Technical Award Winner 💬 "We don't arbitrate the truth. We give people the most powerful tools to make up their own minds." — Chris Bregler, Senior Director / Distinguished Scientist, Google DeepMind; Academy Scientific and Technical Award Winner 💬 “Thinking doesn’t have to be in text! ... We can even use multiple modalities simultaneously to “think” at the right level of abstraction for the problem at hand” — Sergey Levine, Co-Founder, Physical Intelligence; Professor, UC Berkeley 💬“Video is the most general modality that we have that allows us to simulate real-world experience." — Anastasis Germanidis, Co-Founder/Co-CEO, Runway 💬“The relationship between AI and enterprise data is not one-directional. Understanding both sides of that equation is the difference between AI that works and AI that disappoints.” — Dan Roth, Chief AI Scientist, Oracle; Professor, UPenn 💬 "Maybe 99.9% of training data in the next step will be synthetic." — Weizhu Chen, Technical Fellow & CVP, Microsoft AI 💬 "It's maybe the best time ever to start a company — but most 'obvious' AI products will be outcompeted by the frontier labs. The real opportunities lie in solving specific customer problems." — Alfred Lin, General Partner, Sequoia Capital ✨Over two days, we explored one central question: How do we build AI systems that are not only more capable, but also more trustworthy, more secure, and ultimately more beneficial for humanity? This wasn't the end of a conference — it was the beginning of the next chapter for agentic AI. Join us to shape and steward the future of AI for human flourishing! 🙏 A heartfelt thank you to our speakers, sponsors, volunteers, partners, and every attendee (in-person or online) who made this summit possible. 👇 What it your learning, insights, favorite talk, quote, or moment from the summit? We'd love to hear it below!

2/2 🎥 Missed the summit? Watch session recordings online: t.co/pV9eHyxZrt We’re deeply grateful to all of our sponsors and partners for supporting this community and helping bring the Summit to life: @daytonaio,@ema_unlimited, Fetcherr, @LambdaAPI, @replit,@amd, @circle, @databricks, @digitalocean, Novara, @getpostman, @pytorch, @snorkelai, @SUSE, @temporalio, @alice_dot_io, @lobehub, @LudoRobotics, @McKinsey, @nutanix, @OutSystems, @SamsungNext, @vesslai, @trustvijil, @across_ai, @elastic, @GoKiteAI, @neo4j, @novita_labs, @fastinoAI, Valency, @GoogleDeepMind, @lightspeedvp, @MayfieldFund.

Now we are at a critical point. Frontier AI capabilities are advancing rapidly, and the pace of progress is itself accelerating. How do we ensure that agentic AI becomes one of humanity’s greatest advances, while remaining safe, secure, and beneficial for everyone? Join us at the Agentic AI Summit 2026: share knowledge, build an open ecosystem, and steward the future of safe, secure, and beneficial Agentic AI. 🚨 Agentic AI Summit 2026, Aug 1–2 @ UC Berkeley & Online, the largest gathering dedicated to Agentic AI, with ~5,000 in-person attendees & 100,000+ joining online. 👉 Join the livestream now to watch where the future of Agentic AI is going: t.co/pV9eHyxrBV #agenticaisummit2026

6/15 Robotics & World Models The next frontier of Agentic AI isn't just digital; it's physical. What does it take for AI agents to understand and act in the physical world? Can robots build world models that enable reasoning before action? What breakthroughs will make embodied AI truly autonomous? Join Sergey Levine @svlevine, Co-Founder @physical_int and Professor @UCBerkeley; Jim Fan @DrJimFan, Director of Robotics and Distinguished Scientist @nvidia; Michael Spranger, President @SonyAI_global; Anastasis Germanidis @agermanidis, Co-Founder and Co-CEO @runwayml; and Wei Zhan, Chief Scientist @AppliedInt, in this session to learn about how robots and embodied agents can understand, model, and act within the physical world. Moderated by Guru Chahal @guruchahal, Partner @lightspeedvp. 👇 Help shape the conversation in this session: What questions would you like the speakers to tackle? What topics would you like to see covered? Reply below to help guide the discussion!