Built @RossumAi, AlphaGo baseline pachi, bits of git, elinks & other oss... "The world is awful. The world is much better. The world can be much better."

Prague
Joined August 2008
Summarizing my most important tweets from the last two years on consciousness, future, and the narrow winding path through the singularity: pasky.or.cz/singularity/ AI and alignment are the final exam of philosophy.
1
5
1,187
What do people mean when they talk about no thinking traces for Claude? I see the streaming detailed summaries (and often I read them!) just fine. (Yes, they are great.)
The end of detailed summarized thinking traces for Claude and ChatGPT (likely both anti-distillation & for UX) has made being a human in the loop challenging. It pushes us towards automation, rather than augmentation, as it has become much harder to weigh in on in-process AI work
250
My new fav prompt (shopping for a hw project with long lead times): "run Opus+Fable+Argon sideagents to simulate real world process of wiring and hooking everything up and point the most likely fuckups / hiccups / misbuys to identify risks upfront" BTW shout-out to Gemini Argon!
1
1
3
543
(In the pic above, Me = Opus, Best = Fable; just my Pi mode aliases. Argon covers a lot of unique ground! Mixing models *and providers* remains super valuable.) xitter.cf/xpasky/status/21117803…
This /board skill has been working pretty well for me in @badlogicgames's @pidotdev to have multiple models (Fable+Astra+Argon in my case) hash something out: github.com/pasky/pi-agent-co… (Tell your pi to remix it for your multi-agent setup of choice, ootb it's pi-side-agents.)
1
103
My main takeaways: 1. The OpenAI model mix remains as confusing as ever - too many overlaps. Feels like 6.1 Terra is (again!) viable only because there's no 6.1 Luna yet (maybe that's the reason for the stagger?) 2. Models have way too many thinking levels. :)
GPT-6.1 Terra: delivering fast and capable computer use at a lower price. GPT-6.1 Terra understands what’s on screen, takes action across apps, writes naturally, and is a more capable model for your whole workday with the efficiency to take on more of it.
4
1
24
5,813
Petr Baudis retweeted
🥲
The next great "language for agents" should be: - As fast as Rust - As little design space as Go - Have the strictest type/verification system ever written Bend seems like an interesting first effort
59
63
2
3,348
94,259
Petr Baudis retweeted
My quick takeaways on the Anthropic S-1: 1. This is a historical moment 2. $6.6M in revenue per employee!! Hard to even comprehend how good that is but Anthropic is AI native as it gets so makes sense. 3. I always thought OpenAI would drop their S-1 first 4. Customers spend about 9x more after a year. Net Revenue Retention is approximately 900% as of Sep 30, 2026. Good SaaS companies brag about 120%. 5. Cowork is growing faster than Claude Code did. Goes to show you the demand for consumer products. Duo will probably rip. 6. 3 years ago this company did $41M in revenue. In the first six months of 2026, it did $16.3 billion. I don't think anyone has a mental model for that curve yet, I sure don’t. 7. Nearly half of revenue comes through cloud marketplaces. About 48% comes from customers buying through them. Big companies buy where their budget already lives, and the lesson for founders selling to enterprises is to get listed where your buyers already spend. 8. Their run rate formula is published: the last 28 days of usage times 13, plus subscriptions times 12, which i guess is the new ARR math for founders? 9. It looks like They're merging Cowork into Claude dot au which tells you the chat box era is ending and the "go do the work for me" era is here. 10. Thousands of Anthropic employees are about to become liquid at once, so expect the biggest wave of new AI startups and angel checks San Francisco has ever seen. Reminds me when Facebook went public but this is way bigger. Let’s go! 11. Buried in the filing, Anthropic says the gap between what Claude can do and how widely it's used is one of its biggest growth opportunities, so the company itself is telling you the models are ahead of the market and someone has to close that gap for dentists, landlords, and plumbers. 12. Whatever happens, exciting times to be building in AI.
Anthropic has filed an S-1 registration statement with the SEC related to its proposed IPO. Read more: anthropic.com/news/s1-sec Legal disclaimer: A registration statement relating to these securities has been filed with the SEC but has not yet become effective. These securities may not be sold, nor may offers to buy be accepted, prior to the time the registration statement becomes effective. This post does not constitute an offer to sell or the solicitation of an offer to buy these securities, nor shall there be any sale of these securities in any state or jurisdiction in which such offer, solicitation or sale would be unlawful prior to registration or qualification under the securities laws of any such state or jurisdiction.
385
396
87
7,328
808,923
Petr Baudis retweeted
how i work with agents in 2026. ignore all other guides on here.
Replying to @tanvirraj
i tell them what to do. when they do it badly, i yell at them.
59
45
10
889
63,304
the unevenly distributed future
The poll so far between Twitter, Bluesky and Mastodon. I think it looks less absurd than last time, but still pretty clear that the base is divided. xitter.cf/mitsuhiko/status/21143…
7
555
scaling laws for BCI
The human brain is approaching its ChatGPT moment. I spent 6 months writing Phase Lock: a manifesto on how brain computer interfaces can help align AI and preserve human agency. If you only ever read one thing from me, read this.
3
256
china's getting really serious (if this isn't benchmaxxed)
Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model. 🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal. 🔹 Full-project coding: games in Unreal, Unity & Godot. 🔹 Excels at parametric 3D modeling in Blender & CAD. 🔹 More reliable at finance work: Accurate modeling with hallucination down by 65%. 🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks. Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. Open weights by Nov 4, 2026. 🔗 API: platform.kimi.ai 🔗 Tech blog: kimi.ai/blog/kimi-k3-1-pro
5
396
This is actually correct. (tbh for software work just skip Fable altogether, actually for most work unless you need detailed deep real world knowledge; also, Opus high seems just enough to me)
ok listen to me carefully delete all of your ai subscriptions. buy a claude max 20x plan use fable to plan, brainstorm and deliberate on your plans and use opus max to do the work everything else is a distraction and anthropic have won the race, it's over. take advantage of it.
8
62
56,015
who's running this, but for every package in debian?
You’re all prompting wrong
1
1
3
437
oh, maybe anthropic is :) xitter.cf/AnthropicAI/status/211…
New on our Frontier Red Team blog: using Claude to formally verify software at large scale. Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases (or required a lot of expert time). With AI models, however, this is beginning to change. In collaboration with @theoremlabs, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks. Read our research in full here: anthropic.com/research/forma…
1
1
122
yes please, let's make oidc die, such a silly hassle every time
Replacing long lived secrets with short lived secrets (oidc) was a useful idea 2 years ago. I don’t think bearer tokens with 15 minute expiration are gonna stop you from getting owned. Proof of possession (e.g. TLS) is the only way. OIDC will be dead in 2 months
254
Petr Baudis retweeted
Ronald Reagan eats the same Big Mac as you Tim Cook uses the same iPhone as you the NSA runs the same abliterated GLM-5.3 as you
44
192
17
4,985
107,769
I'm really enjoying watching this one. Hope it scales!
WE DID IT! We can now pretrain transformers without backprop and even beat backprop+Adam on validation loss with insane speeds. Now the big question is scale. It’s Ironic, we have reached a breakthrough In AI capability due to inspiration from a safety blog
2
354
It is insane how many TUIs are reimplementing their own poor laggy mouse clipboard control, just because there is no way to define "mouse selection regions". That solution seems pretty obvious. Is it somehow harder than it seems? @mitchellh is someone thinking about that yet?
1
261
claude> The gap: the patched PHP 8.4.27 installed automatically at 16:00, but nothing restarted Apache, so the sites kept running the old vulnerable PHP until I restarted it at 21:39. ...really, Debian?! :pepehands:
2
323
turns out Erdos-Strauss is a really hard problem :) (I have had a swarm of a variety of models working on it on/off for 2 months now, no budge yet)
oh my god, are we *really* going for this now, fable? the breaks are getting a bit out of hand
1
3
342
Please use the robot safely and in a friendly manner.
Unitree Introducing: H2-X Kung Fu Legend Guardian🤖 Powerful performance to serve and guard you, with a maximum endurance of approximately 8 hours. Please use the robot safely and in a friendly manner.
10
2,104