So many people have DM’d me about this trailer and I’ll be honest, watching it made me physically throw up. I was COO from before FTX even had a name, it was my baby, and I was the last one out of the Bahamas after Sam got extradited. I lost everything I had on it and it took me a long time to get over, so I have zero interest in defending the people who caused this. The media loves the story that FTX blew up because we were all partying in the Bahamas, but the place I worked looked nothing like that. We had under 200 people in the core company at its peak, not counting subsidiaries, when other exchanges our size had thousands, so a lot of us were doing 16 to 18 hour days for over three years and barely had time to sleep, and whenever we walked in, day or night, Sam was at his desk. I worked with Caroline and never once saw her get aggressive like she does in this trailer. She’s quiet and mild, and often lost in her own thoughts. Most of us didn’t even know she and Sam were dating until we read about it after the bankruptcy, and I never saw him so much as put an arm around anyone. FTX’s collapse ruined so many people’s lives, including ours, and Netflix decided to make it a love story. Shame on you @netflix.

@Muse learned a new trick tonight: email triage with Jev, TypeSafe’s System One model. For now, it’s still in eval mode. The setup is simple: Jev takes two things, the state (the email: sender, subject, body) and the criteria (my written definitions for each category). Emails are sorted into four categories: needs action, newsletter, worth a glance, and noise. Every email gets classified twice, once by Muse and once by Jev. I only review the disagreements, and my decisions become the ground truth. Then I rewrite Jev’s category criteria and re-score.

AI is already smart enough. Real-world tasks are still slow, expensive and unreliable, because we hand AI a computer built for humans, then wrap it in a heavy harness. AI doesn’t need to get smarter. It needs a computer built for it. Today we’re releasing Pine Computer

launched a model accidentally served trillions of tokens a day 29.4% of the fortune 500 showed up raised a really big series A from @a16z it's been 3 weeks we would like to sleep now

A lot of hype around OpenAI's Astra model here on my timeline today. Apparently, this goes back to a new article from The Information, which said Astra is a "recurrent depth or looped transformer". It's always interesting to read about new or different approaches (including rumors about what the closed labs may be up to), but let's debunk this a bit. About 2 months ago, I shared the architecture details of Nanbeige, for example, where "Nanbeige4.2-3B is pretrained from scratch on 28T tokens with a Looped Transformer that reuses the layer stack to increase capacity without adding parameters." Yes, that's it. The looped transformer idea is just reusing layers in the transformer block. In the case of Nanbeige, the main idea is to reuse the same 22-layer stack (=transformer block) twice instead of once. So, effectively it extends the 22-layer architecture to 44 layers, but without duplicating the weights. In simple terms, this roughly doubles the size of the model (if we ignore the embedding and output layers for a second). But instead of requiring 2x the storage and RAM to host this model, it stays at the same size since we reuse the components. However, it's almost 2x as expensive in terms of compute, because we run the embedded text through almost 2x as many layers. Why? In the Nanbeige 4.2 technical report, the researchers found that two passes gave the best trade-off and retained about 75% of the token efficiency of a standard architecture. (More passes gave barely any gains but made the training much slower and much more expensive.) While, as far as I know, Nanbeige 4.2 is the first notable open-weight model that adopted this approach, the idea goes back to the NeurIPS paper "Mixture-of-recursions: Learning dynamic recursive depths for adaptive token-level computation". Actually, this paper proposes a mechanism that is a bit more sophisticated by adding a learned router that determines whether each token receives one, two, or more passes. So, easy tokens can exit early while harder tokens receive additional computation. In sum, Astra may be a really good model, but this shouldn't be about this "looped transformer aspect," which is just a tiny architectural tweak. Also, the statement "the new technique works in a way that obscures some or all of the AI's reasoning, otherwise known as 'chain-of-thought'" is not necessarily true with respect to the looped transformer method. It's possible that The Information journalist refers to some other technique or misunderstood the looped transformer method. Reusing layers does not by itself suppress visible chain of thought. It adds computation in hidden states before the next token is emitted, just as ordinary transformer layers do. But based on the information we have, the only plausible interpretation here is that if a model uses more of these recurrent passes, it may need to generate fewer intermediate reasoning tokens. So then more of its computation happens in latent activations that cannot be read as text. But we would get the same effect if we were scaling up the model size, like GPT 5.6 Luna -> GPT 5.6 Sol.

DeepSeek Harness 发布短短两天，GitHub Star 已经突破 10 万。 它已经远远超过存在时间更长的 Qwen Code，也超过了非常受欢迎的 Pi，甚至已经超过 OpenAI Codex。 目前前面只剩 OpenCode。 更夸张的是，这些项目大多已经积累了几个月甚至更长时间，而 DeepSeek Harness 只用了两天。 以目前这个势头，超过 OpenCode，恐怕也是指日可待。

Two months ago I was testing Grok @imagine to make the Battle of Red Cliffs (赤壁之戰). Now in Grok @bot you can spin up a director, screenwriter, music producer, and cinematographer to create a cinematic movie trailer in one shot. Attached the only prompt I gave to Grok @bot and the rest was autonomous.

We’re ending our partnership with Cursor following its acquisition by SpaceX. Under our proposal, Cursor’s direct access to our models would end on November 12. We know that the people most affected by this decision are the developers who rely on OpenAI models in Cursor. We care about their experience in this transition and we’re ready to go above and beyond to support them.