we've been working hard on the best way to collaborate across agent sessions - across any combination of dev workstations, mac minis, or cloud environments. Send prompts to your teammates sessions teleport worktree bundles + task context across workstations manage sessions and collab on artifacts from your phone what else should we add?

>may be really civilizationally useful in giving us time to adapt to what might be a fast takeoff. Agree - this has really interesting implications on RSI, which is that even if you had Noam Shazeer level models tomorrow they'd still be experiment bottlenecked, so it changes the shape of takeoff. very important factor to model. >But my argument would be the same: rather than worrying about inequality we should focus on growing the pie so fast that even the bottom of the income/compute distribution are much better off here in America than almost anywhere else. 1000% agree. My median expectation is that compute spend follows the 2-3x yoy trendline (so that next year it is >2T), and that a given level of intelligence keeps getting 10x cheaper every year due to efficiency improvements throughout the stack which will hopefully compound into intelligence too cheap to meter. Will take you up on that dinner in Boston sometime, or let me know next time you're in SF! And don't worry - I drink, can't take the Australian out of the guy

to my knowledge this is the largest open experiment on autonomous agents iterating on a research environment we scaled runtime, compute, diversity of models and harnesses. as a comparison, similar tasks on oai/anthropic system cards are anthropic "optimizing an llm training on CPU" and openai gpt 5.6 doing nanogpt track 1 but for less than a day. we also share a lot of the details (traces, scratchpads, ect..) so you can look into how models approach such tasks this experiment is quite noisy, one run in the same setting has a ~50 step spread after 24h. i find it super impressive that while models explore relatively the same ideas and the task and environment have a lot of variance, there is still a big gap between different models fable 5 closed 82% of the gap to the current human record with kimi K3 being very impressive as well. i'm currently running grok 4.6, deepseek v4 pro, muse spark 1.2, qwen 3.8 max and glm 5.3, expect results next week one of my favorite parts is some of the tools that models developed during this experiment, especially with prime agent. for instance kimi K3 created its own experiment API for generating optimizer variants, loss comparisons, Newton Schulz tuning etc.. we also have an early deepseek v4 pro <> prime agent run that did a PSGD experimentation outside the normal nanogpt loop to build intuition before starting gpu runs. other cool examples in the blog with other harnesses as well! i'm also very excited about the ideas we have in mind on this subject, we will keep working on understanding the research capabilities of (closed and open) frontier models

We ran the largest open experiment on how frontier models do AI research. 100+ autonomous runs across 10+ models, sandboxed on 8xH200s for up to 8 days, iterating on the nanoGPT optimizer track. Best runs closed 82% of the gap to a record built by dozens of humans over months.

My stubborn and completely uninformed opinions on AGI: - when AGI finally arrives it will be blindingly obvious and there will be no debate whether it’s here - this is in fact a good proxy test: so long as reasonable people earnestly disagree, it is not here. This test is circular, unsatisfactory, yet useful - we will have a precise definition of what AGI is when the proxy test is passed. The reference implementation will inform the definition - we have not passed the Turing test. People were tricked for a while the way they were with Eliza, but that decays fast. Not only can you tell you’re chatting with a model, you can often tell which model it is (of course the level of sophistication is incomparable) - this says nothing about the timelines. Labs may have AGI tech internally already, or it may take another few years of doubling, or we may be far away, I don't know - none of this diminishes the risk or utility of what's coming

I run 20 coding agents in parallel as my everyday workflow. Today, I’m launching Jcode. It’s an open-source agent 20x more memory-efficient than Claude Code, so you can run 20x more agents at once. Try it today:

We re-tested GPT-5.6 Luna from @OpenAI on ARC-AGI (Verified) following its recent 80% price reduction: - ARC-AGI-2: 59.6%, $0.18/task - ARC-AGI-1: 90.7%, $0.07/task The new results match Luna's original performance at a much lower cost.

My summer project is done! A 20 video, free course on post-training to accompany my book is all on YouTube with slides open for modification & re-use. ~12 hours of content covers the core foundations and some research areas I think will grow in importance. It was a fun time to review all the fundamentals again, as it is clear in the next 1-3 people the amount of people wanting to learn post training will likely 100X again from today, as we have already 100X'ed from two years ago. As AI agents get increasingly capable at coding and discussing these fundamentals (see the code exercises accompanying the book that I am refining with the community) I think developing clear intuitions for how models work and why is one of the most important skills going forward in AI. Still, learning the post-training math is the best way to battle test them. I personally just in this course am starting to master how forward/reverse KL relates to post-training topics. Thanks to all my viewers, and I'm happy to answer questions in the book discord or understand how to better teach the various reward models, on-policy distillation, new RL algorithms, etc. Plus, the book is 50% off right now with the code PBLambert on Manning to celebrate the launch. I'll share the relevant links below. Who's going to make this course for pretraining?