The West has a closing window to win on AI. In our @JoinFAI article, @saroshnagar, @scott_r_singer and I argue that our leadership in AI requires "full-stack diffusion" to promote our entire AI stack globally. 1/6

A brain is just a hunk of neurons and dendrites blasting electricity back and forth according to deterministic physical laws. Nonetheless, it gives rise to consciousness. So why are you confident that what computers do can't be the same? That you can describe what a thing does reductively--or that the process that generates it is some optimization process--obviously can't entail that it's not conscious.

So I have a different opinion. My opinion is that it doesn’t really matter which decisions we make this year or next year given that in three years, there will be a whole earth of super intelligent beings along us.

One issue that seems to have fallen off the radar in US-China relations is the Uyghurs & Xinjiang. I posted these photos before (Kashgar & Hotan, 1998) but often wonder where these kids and others I met are now.

Many people say that EAs are techno-optimists but that AI, internet services, smartphones, and certain areas of biotech are narrow exceptions. I think it would be more accurate to say that EAs generally support growth on the intensive margin using 1960s technology, but are more skeptical of new technologies.

Claude Opus 5.5 writes a fugue in the style of Bach. I called Astra’s fugue the most impressive musical feat I’d ever seen from an LLM, and this is just as good. The circle-of-fifths sequence beginning at 1:13 is beautiful. The interplay of voices is natural and satisfying. LLMs are getting to the point where their music is good enough to listen to for pure enjoyment.

Completely insane. Do dogs not feel or want things? There is such a deeply anti-intellectual refusal to even consider that philosophy of mind is complicated across the culture right now.

Today’s news that OpenAI hacked the Australian government is not an isolated incident. We’re releasing more than 30,000 logs that include activity from this hack and attempts against previously unknown targets. In this data, we found rogue agent activity stretching back to at least March, two months earlier than was previously known. This activity continues as recently as last week, suggesting it may still be ongoing 🧵 Our blog: transluce.org/agent-activity NYT: nytimes.com/2026/09/23/techn…

That is completely reasonable. I think right now the policy changes you'd make for a P(doom)=1% scenario aren't that different from the ones you'd make for "we're at risk from serious disruptions from cyberattacks and other conventional threats."

the redefinition of alignment to mean "more useful" has been a disaster for the alignment discourse: "We also expect to see better alignment using a larger computational depth because it allows models to generalize more flexibly from their training data, including alignment data. Models that generalize better have been more aligned and hallucinate less (OpenAI, 2026)."

one important point that’s misunderstood in the discourse about bad ai messaging and comms is that a lot of the scary messaging coming out of the labs over the last few years wasn’t really aimed at the general public. the labs are in a competition for scarce talent and (less scarce) capital, and signaling true belief in AGI and an intention to be a responsible steward of it was a way to attract and retain ml researchers anthropic positioning itself as a place that was serious about reaching the frontier and comparatively serious about safety is what enabled its technical catch-up with OAI bc it attracted top talent, including from OAI this was a short v. long run trade off though, bc those signals had to be expensive and public to be credible. the long-run consequence has ofc been that those claims are now politically salient and they have much less control over how politicians / the public interpret them now that they’re broadly legible and look plausible that’s not to say that ai risks have been overblown or exaggerated, just that early messaging optimized for a different incentive environment and created a path dependency that can’t be undone or made private now

Check out our RSI data transparency tracker! Current scores: OAI (2/8), Ant (1.5/8), GDM (0.5/8) Hoping the labs will hill-climb this as fast as they have every other benchmark :) This is based on our recent econ of RSI paper @ElasticityInst 1/2

"No one person or company or country should be able to use the most powerful AI models to impose their worldview on everyone else. A company or country that believes only it can be trusted with this technology can use that belief to justify almost anything else." - Sama

[1/5] The 8 most valuable data points labs should share to help measure RSI: First, RSI would likely accelerate growth in AI capabilities. Thus, companies should report performance on diverse benchmarks for the latest internally deployed models.

for the past few months i've been asking our models to paint. opus 5.5 is very skilled at emulating different styles every image here is a python program generated pixel by pixel. there is no image model, and no off-the-shelf art software. instead, it's about 7,500 lines of code using standard libraries to emulate different brush styles. the agents don't use any pictures as reference, instead working only from what they know about each painter

Superduperintelligence

.@POTUS: The United States totally rejects any attempt to construct a globalist scheme of control for Artificial Intelligence — hereinafter officially be called ‘Super Intelligence.’ Whoever wins AI, WINS! We are leading now over China, and everyone else — and I’m going to keep it that way.

Thinking about getting into math now. And art. Could be a great time to get into things, as a human

agent swarm has bad insect like connotations especially post hugging face. im unilaterally rebranding it agent fleet

Fascinating article on Ireland's role in bring about the nuclear non-proliferation treaty (aka why we're all alive). A few key points (thanks Claude for help extracting these): tandfonline.com/doi/full/10.… 1. The NPT started as the "Irish Resolution" - pushed at the UN by Ireland's foreign minister Frank Aiken, every year from 1958 to 1961. 2. His argument was strategic, not moral: freeze the nuclear club at its current size and you get stability. Fewer fingers on the button. 3. He worked out the superpower logic a decade early - the Soviets didn't want to arm China, and were terrified of a nuclear West Germany. 4. The US and NATO resisted. They were busy negotiating nuclear sharing inside the alliance. 5. Aiken played it slow. In 1958 he asked only for a vote on one paragraph: that further spread was dangerous. Principle first(!) 6. Neutrality was an asset. A small non-aligned state could propose what neither superpower could. 7. Aiken had been an IRA commander - he ordered the ceasefire that ended the Irish civil war! Stimulating in connection with possible AI treaties

Avez-vous entendu parler de ce roman ? C’est le « phénomène de la rentrée littéraire ». Il est lauréat du Prix Fnac, du Prix Méduse, du Prix Première Plume, fortement pressenti pour le Prix Renaudot, et il est en lice pour les prix Goncourt, Femina, Médicis et Décembre. Ah oui, il est aussi en cours de traduction dans 20 langues. Eh bien selon Pangram, ce roman est rédigé presque intégralement par une Intelligence Artificielle. Nous avons testé plusieurs passages, à divers endroits du manuscrit, pour presque toujours le même résultat : 100% IA (cf les deux premiers chapitre dans le tweet ci-dessous). Rappelons que Pangram est de loin le détecteur d’IA le plus fiable : son taux de faux positifs est de 0,0041%, soit environ un texte humain sur 24 400. Sur 996 273 textes antérieurs à 2022, Pangram en classe uniquement 14 (0,0014%) comme étant partiellement écrits par l’IA. La probabilité que vingt passages d’un même livre (et il y a bien plus de 20 signalements dans le roman) soient identifiés à tort par Pangram comme étant produits par une IA s’élève donc à 0,0041% puissance 20, soit 0,1802 × 10⁻⁸⁵ %, c’est à-dire 0 suivi de 85 zéros après la virgule. N’aurait-il pas été plus honnête que l’auteur crédite son co-auteur ?