AI data center buildouts already rival the Manhattan Project in scale, but there’s little public info about them. So we spent the last few months reading legal permits, staring at satellite images, and scouring news sources. Here’s what you need to know. 🧵

Ericsson & Simon (1980) takes the middle cell psycnet.apa.org/record/1980-... Now, your nominations for minimal citation impact but moderate actual impact? A contribution that altered policy or practice... A paper that extinguished an active area of research... An uncredited gem...

At #NeurIPS2025? Join us for the San Diego Alignment Workshop: Open Social, an evening of connection & conversation hosted by FAR.AI. 📍 Omni San Diego Hotel ⏰ Mon Dec 1 | 7–9PM PST RSVP by Nov 30 👇

One thing I appreciate about social media is that people who have a bee in their bonnet about one specific niche thing can reach my attention. See also @andymasley.bsky.social on the water usage of AI.

Very well done story on 60 Minutes about the harms of grant terminations and freezes at Harvard with Joan Brugge, Don Engbar, David Liu, and a very compelling young cancer patient, now cured with Liu's technology. Transcript and video here www.cbsnews.com/news/researc...

Choose carefully what you optimize for.

That Watson was a disagreeable guy should make you more dubious of stories that minimize his contribution, not less.

This is probably the most literate discussion of alignment problems across crypto and AI protocols you’ll find anywhere that’s still generalist accessible. Emmett Shear and Alex Stokes really went at it with some high-level and sophisticated sparring. www.youtube.com/watch?v=zHGg...

The concept of cached assumptions is good to have in your mental toolbox.

This is a neat series of posts.

We’ve been heads down @ArcadiaScience for a bit but the @arenabioworks news this week caused me to dump thoughts. hard things are hard; don’t be such a fucking hater. Reflections on parallels w our own institutional experiment here seemay.substack.com/p/big-experi...

Provenance! How do we know who runs a website? Digital signatures. Content can be signed too. Doesn't mean unsigned content can't be created, but browsers signal signatures via padlock and the same thing can happen for content.

Yes, liability is the obvious approach here. It's how mature industries open up opportunities for innovation & reduce the need for regulation.

Excited to launch an openRxiv partnership with the scientist-run AI review service qed (@qedscience.bsky.social), the brainchild of @odedrechavi.bsky.social 1/n openrxiv.org/enabling-rev...

"Even with notable gains in fluency, GPT-5 is still prone to hallucinate, break rules, and exhibit latent capabilities that raise safety and biosecurity concerns" www.nature.com/articles/s41...

Great post this week from @lisalibrarian.bsky.social that hopes to clear up some of the confusions around Creative Commons licenses and the use of such content for AI training. This speaks to the ongoing failure of the publishing and OA communities to make clear just what these licenses mean

When @fermatslibrary.bsky.social brought up this 1940 article about why we have nothing to worry about from nuclear chain reactions, I checked that it was real and not a modern forgery. Because it seems almost too good to be true in light of current AI safety talk.

🤯

A network of peer reviewers in Italy is targeting medical journals, threatening “both the scientific record and patient safety,” a team of researchers including @deevybee.bsky.social report.

I’m going to run a conference called “HolyShitCon🤯 2026” to talk about AI developments. Who wants to come?

It's fun to try, but, like with spam, we need several systematic interventions: blacklisting producers & statistical techniques will only do so much. Ultimately we'll need to not trust by default - anything that wasn't produced by a member of a chain of trust is suspect.

So important given what this administration is doing to US Agencies.

Announcing our Frontier Data Centers Hub! The world is about to see multiple 1 GW+ AI data centers. We mapped their construction using satellite imagery, permits & public sources — releasing everything for free, including commissioned satellite images. Highlights in thread!

Europe has a chance to shape a safer and more values-aligned future for AI innovation, and needs to. This was my main message at the AI in Science Summit in Copenhagen today. I also presented Scientist AI, LawZero's approach to create technical guardrails and help accelerate scientific discovery.

🤯

𝗧𝗵𝗲 𝗘𝗨 𝗔𝗜 𝗔𝗰𝘁 𝗡𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿 #𝟴𝟵: 𝗔𝗜 𝗦𝘁𝗮𝗻𝗱𝗮𝗿𝗱𝘀 𝗔𝗰𝗰𝗲𝗹𝗲𝗿𝗮𝘁𝗶𝗼𝗻 𝗨𝗽𝗱𝗮𝘁𝗲𝘀 is live! CEN and CENELEC have announced exceptional measures to speed up the development of European standards supporting the AI Act.

AT service operators and moderation thinkers: I put together an early proposal around infra abuse notices across organizational boundaries. really looking for feedback on this one, it is bait for counter-proposals and references to prior work!

I'm putting a sign on my stove for Known Hot November.

Solving the blank page activation energy problem is a great use case.

I'll just put this out there. There's a reason we have representative democracy and not direct democracy in the US. Blocklists are direct democracy. Perhaps different atproto servers should be able to elect moderators like we elect representatives?

This labeler is such a clever way of making spam visible. Bluesky wins yet again: bsky.app/profile/stec...

This dystopia is brought to you by a16z, the firm that is against ethics.

I love asking the question: how much should be spent on scientific replication? When asked, I have suggested that 1-2% of a funding budget felt right. "Felt right" meaning no evidence basis other than intuition. @jdworkin.bsky.social offers a much more thoughtful approach.

We now finally have actual semantic search, which is huge.

”If Anyone Builds It, Everyone Dies” was recently added to the New Yorker's “The Best Books of the Year So Far” list! newyorker.com/best-books-2...

You Can Just Do Things lite.cnn.com/2025/10/30/t...

You could do worse than be a normie.

Triangulating who someone really is by isolated statements they make is the primary activity of a lot of people here and the lack of epistemic rigor is really telling. 🤔

We estimate the average ACA enrollee will see their out-of-pocket premium increase by 114%. But, it will vary a lot based on age, income, family size, and zip code. I'd be interested in what you're seeing if you're an ACA enrollee and log on to the marketplace. Please reply here.

The conversation here should be "To win, we need to not push our specific issues into the national spotlight so much" rather than "Haha look this thing I don't like is unpopular"

We've launched a new tool to track AI progress! The tool addresses one of the field's biggest challenges: benchmark saturation. It's called the Epoch Capabilities Index (ECI) — here's what makes it different:

AI brings both promise and peril to defense. Steve Kelly addresses 3 questions: What's its military value? Does it stabilize or destabilize? Will AI become the threat? His concerns span cognitive decline from over-reliance to unexpected emergent behaviors in multi-agent systems. 👇

Transparency is good and companies should not feel hesitant to share bad news about how people are using their products. This is definitely bad news.

Layoffs are contagious. www.cnbc.com/2025/10/28/a...

GOOD QUESTION: "Why are universities still teaching referencing styles?" www.timeshighereducation.com/depth/why-ar... (paywalled). I have seen first hand the stress this causes students! Article is a broader discussion about referencing styles and what a waste of time they are.

Transparency is important, even if it leads people to disagree with you.

Why might malicious actors choose less lethal but easier-to-hide weapons? Olivia Shoemaker argues current AI evaluations focus on scientific capabilities, often missing operational factors that determine real-world misuse patterns like evading detection and acquiring resources.👇

New: a $60 mod to Meta's Ray-Ban glasses disables the privacy LED light. This is supposed to light when people are filming with the glasses. We bought the mod, verified it works. Now you can never be sure whether someone wearing Meta Ray-Bans is filming you or not www.404media.co/how-to-disab...

What's a guy gotta do to get some frickin sharks with laser beams around here? lite.cnn.com/2025/10/23/b...

Some things work best offline: gossip and normative discussions are two.