Two things to distinguish: Did any human or agent look at user data as part of the Navier Stokes effort? No. Do we use user feedback and de-identified data to improve ChatGPT and Codex in a holistic way? Yes. And so does every LLM company.

We solved Navier-Stokes! Many of my colleagues left their fields because they believed that working on AI would be the fastest way to solve their fields' grand challenges. It's surreal to see the first of those challenges fall and to watch a future we believed in become real.

Agree with @JensenHuang: we’re entering the AGI era. The AGI era must also be the alignment era. We need to teach AI to love humanity and train AI monitors as capable as the AIs they supervise. Said best by @merettm in this thoughtful, sobering piece:

Alongside the launch of GPT-6 Astra, we're committing $1 billion to subsidize Daybreak access and frontier capabilities to help frontline defenders protect essential services and critical infrastructure. Welcome to the AGI era for cybersecurity.

My biggest wow moment using GPT 6 Astra had nothing to do with the visual posts you've been seeing... Don't get me wrong, those are awesome and fun... But the biggest wow moment was when I gave Astra context over everything about my business and then gave it the prompt below. Email, Slack, Business Texts, Notion, and every meeting note... everything I've been doing (Which involves like 20 different projects with many teams over the past few months)... Here's the prompt I used which was not really well thought out just a wispr flow ramble... All I can say is the response to this prompt was the most useful thing an AI has ever given me... ""I'm about to plan the next month, quarter, and year" Make one document that aligns me on all of those timeframes. Look for my flaws, look for my strengths, and highlight those. Look at ALL of my activities, which activities do you think i'm wasting the most time. Which activities should I do more of? Of the people I work with who seem the most dependable? Where am I the least dependable? If I could only upskill in one area over the next year what would it be and why? Where can I organize my company better, and how would I do that? Use text when necessary, use charts/visuals when necessary, but never ever fill space for the sake of it. Every graphic and word should matter. I want to know about finances, relationships, business model, everything. ""

A concerningly common take seems to be that keeping Chain of Thought monitorable doesn't matter because interpretability will save us, or it's already useless This is total bullshit. CoT is our best current tool for safety & interpretability, losing it would be a major tragedy

GPT-6 Astra recreated the Palace of Fine arts in Blender. This is favorite building in San Francisco because it was built for the World's Fair in 1914 where the steam locomotive and telephone were showed off. It was a time when technology gave us all a deep sense of optimism for the future. I feel like some of that sense has been lost since, but I hope models like astra can help restore it, and push us towards a more hopeful future.

GPT-6 Astra is here! This is a big moment for our research team - years of work on pretraining, reinforcement learning, and post-training have come together in our most capable and aligned model yet. It can build and test software, work across apps on your computer, and even help you take a crack at open scientific problems! Capabilities that felt like grand challenges a few years ago have become tools people can actually use. One example is Computer Use - if you’ve tried this before and felt like it was too slow or not good enough, I encourage you to give it another shot. We’ve come a long way since Operator, and it “just works” now. We’re also asking these systems to act on your behalf for more consequential work. Agents needs to stay aligned with your goals and values, think transparently, and respond to oversight even when tasks become difficult. We’ve made substantial progress on these behaviors in Astra, alongside stronger monitoring that can stop potentially unauthorized actions. That work is part of what makes this release possible. I think alignment is one of the most important research frontiers in AI, and it remains far from solved. Our ability to understand and align models has to keep pace with model capabilities. We want to give people more room to think, build, and discover with increasingly powerful tools that remain *under their control*. Huge thanks to the researchers and teams who got us here. There’s a lot more work ahead, and I’m incredibly excited about what we can make possible in the near future!

We temporarily slowed some frontier training to strengthen security and monitoring. Our largest planned frontier RL run remains on hold while smaller-scale training and evaluations help us test safeguards and gather more evidence of alignment. I expect confidence in safety to increasingly set the pace of AI development. We urgently need tools for labs and countries to coordinate on this, which is why I signed Pacing the Frontier. In the meantime, we’re taking practical steps ourselves - and will continue to share what we learn as our approach evolves.

Many of our strongest researchers are choosing to focus on alignment, but we're also hiring! If you want to work at a frontier lab which takes alignment seriously and doesn't pretend it's solved, please apply.

WOW!! 🤯 Among many jaw-dropping results, this proves NP-hardness of the Closest Vector and Nearest Codeword Problems for *polynomial* approximation factors, for the first time ever, and via a totally new approach (Reed-Solomon techniques). Amazing!

yes, nonsofic groups exist: this statement is one of many new beautiful results proved by Astra, our next major model. We're releasing 10 such Astra proofs, complete with lean certificates and CoT walkthroughs for each of them. The results are wide-ranging, from von Neumann algebras (disproof of Connes' Rigidity Conjecture) to better bounds for high dimensional sphere packing, for circuit complexity, for monochromatic triangles in multicolored graphs, and more. More thoughts here:

This one is personal for me: 4 years ago I decided to go into AI development because I thought there was a chance that by the end of the decade AI could be better than me at math: well it happened ahead of schedule, 2026 instead of 2030 😅. It is 100% clear that science is being fundamentally transformed before our eyes, and it is NOT aspirational anymore to say that AI will accelerate science. But this will only happen if scientists can actually use SOTA models. Science is best done by scientists. The new ChatGPT for Academics is designed exactly for this purpose: we want to empower our users and not hold this power for ourselves. I can't wait to see where the frontier of knowledge will be very soon, we have so many questions we want answers to!!!