We’ve collaborated with creators on producing 14 videos. Their audiences total over 20 million subscribers. We’re doing this because most people are still in the dark about the risk of extinction posed by superintelligent AI.

We need humans not just "in the loop", but in the design of these AI systems that increasingly affect all our lives. Even with perfect performance, we need transparency to build trust. www.nytimes.com/2026/04/15/m...

This new report from Stanford shows quite how unpopular AI is becoming. Some highlights: - 64% of U.S. adults think it will lead to fewer jobs; only 5% think it will lead to more jobs - In the UK, only one in five people thinks AI will improve the job market 🧵 1/2

Vacation rentals, hospitality not included.

ServiceNow leading the charge to replace workers with AI. www.theregister.com/2026/04/13/s...

New Fed paper by Crane & Soto uses official labor force survey data and finds that 500,000 fewer coders are working than pre-LLM trends would predict. Hits entry level, not senior workers. @jburnmurdoch.ft.com piece: www.ft.com/content/b69f...

Asking Professor Stuart Russell: Do you think we should pause the development of AI?

Labor and Capital have co-existed because they need each other. Superintelligent AI threatens to throw off the balance by relieving Capital of its dependence on Labor and neutralizing the power of strikes. However, we can still boycott. www.theguardian.com/commentisfre...

AI is changing our jobs: among people who use AI regularly at work, 27% say AI has replaced some of their tasks; 21% say it has enabled new tasks. This and other workplace usage findings from our new Epoch AI/Ipsos survey on AI usage in 🧵

The Pentagon’s race to adopt AI without effective regulation poses serious — even deadly — risks. Congress must establish guardrails. Learn more in our report on the military and AI: bit.ly/4sJCA2S

The face of the Pentagon's punishment of Anthropic held xAI stock and sold it for millions www.theguardian.com/us-news/2026...

The whole "develop a dangerously capable AI in private and selectively release it" approach that Anthropic is taking with Mythos is not likely to work sustainably, and once the weights are leaked, the damage is irreversible. lite.cnn.com/2026/04/08/c...

Happy to see this

The #Trump defunding of #usaid could lead to the shutdown EM-DAT, the world's largest database of natural and technological disasters, maintained by the #ulouvain. https://statsandr.com/blog/em-dat-the-world-s-disaster-memory-is-at-risk/ #belgium #data #defendresearch #emdat #funding #opendata […]

None of this is hard. But you have to CARE enough to actually do it. I wrote a full guide 👇 www.dsebastien.net/slopsquatti...

Why would ScaleAI, a company with Pentagon contracts, be doing this particular task? www.theguardian.com/technology/2...

Some new info in here, even for people well - briefed on OpenAI history. www.newyorker.com/magazine/202... Yes, people quitting will affect how reckless OpenAI chooses to be.

The Case For Doing Nothing nosidebar.com/doing-nothing/

OMG it's just a PDF on a website how much are the fatcat Arxiv execs raking in on the backs of our volunteer labor!?!!

Some good signs for No Kings!

CALL TO ACTION | Governments must act now to manage AI for the public good Elders and leading Artificial Intelligence experts urge governments to tackle escalating AI risks to ensure technology serves humanity, rather than endangering it. Read in full: theelders.org/news/governm...

Led by @kennypeng.bsky.social, we've been working on a new implementation of an old vision, to go beyond "feeds" to "interconnected trails". Demo on Bluesky/ATProto posts over a week! www.skytrails.org

www.theatlantic.com/magazine/202... Good article about the dangers of trusting "almost perfect" technology ⬇️

What's up with the political positioning of the AI companies lately? "It's not (just) about money, it's about power", says David Krueger of Evitable. www.salon.com/2026/03/23/w...

Cybersecurity is one of AI's biggest risk domains—Attackers only need one exploit, but defenders must patch everything. Dawn Song reveals frontier AI can find vulnerabilities for just a few dollars and argues for a shift to formally verified, secure-by-design systems.👇

AI has solved one of the problems in FrontierMath: Open Problems, our benchmark of real research problems that mathematicians have tried and failed to solve. See thread for more.

WPI is currently seeking individual donors and established nonprofit partners working to advance biomedical research who are interested in piloting data-guided decision-making. Email us to connect, info at woodleyparkinstitute dot org

Makes you wonder what they're up to now: lite.cnn.com/2026/03/20/u...

To watch this event either live or later, tune into open.substack.com/live-stream/... with @angierasmussen.bsky.social and others previewing the Ridley lecture, providing live fact checking, and discussing it afterwards. 3/3

More recordings from the London Alignment Workshop are now available. New videos from Neel Nanda, Marius Hobbhahn, Robert Trager, Sören Mindermann, Ryan Lowe, @scasper.bsky.social, @conitzer.bsky.social, @zhijingjin.bsky.social + more. 👇

The results were inconsistent and often underwhelming. Even the best in class – LinkedIn and Pinterest – only labeled 67% of the AI content I posted. YouTube managed roughly 50% and TikTok about one third. Instagram did worst of all, labeling just 15 of the 105 synthetic images.

This was a fun story and a very plausible future: nearzero.software/p/warranty-v...

Do you have any thoughts on systemic bottlenecks in science and how to fix them? Astera is running an essay competition! astera.org/essay-compet...

Some of our best grants have been in talent, helping people find their way to high-impact careers. We think there's way more room to grow here, so we're launching an open RFP. If you run an org that helps people have an impact with their career, apply! coefficientgiving.org/funds/effec...

Today we report single-cell APEX-seq (scAPEX-seq)—a method for unbiased mapping of subcellular transcriptomes at single-cell resolution. It reveals cell states invisible to standard scRNA-seq and identifies regulators of CAR T function that improve solid tumor killing. tinyurl.com/32pf6b8p

If you are working with multilevel data, try out my new package mlstats 📦 — an #rstats package for multilevel analyses. You can create publication-ready descriptive tables like the one below (including within- AND between-group correlations + ICCs). github.com/felixdidi/ml... Details in thread 🧵↓

🚨 Want to work at @nature.com with me? We have an ultra-rare opening on my team of reporters. This reporter, based in NYC or DC, would cover physical sciences, energy, environment, Al and policy. You'd get to work with some of the most talented editors in the biz! DM w/ any Qs. Apply by March 20!

Hot take, but the court shouldn't be the final authority on matters of science.

Update 3/16: A very small handful of NIH apportionments have come through — with big exceptions. The ones that OMB has approved are all Congressionally mandated/earmarked. The broad discretionary funds that fund most NIH grants not yet been authorized. (1/2)

I've got a joke about randomized controlled trials, but only some of you will get it.

Totally normal technology, just like computers and the Internet, right? www.bbc.com/news/article...

This is an incredible documentation of the systematic, intentional collapse of American health science, by Elizabeth Ginexi. open.substack.com/pub/elizabet...

Funny to think this was only about 12 years ago, now both of those things are trivial. What would be the "research team and 5 years" task today?

Speaking of regular expressions, it's cute how Claude Code is always trying to write them when it's processing smallish amounts of data. I'm like, "You're a LLM, just read everything into context and extract it."

The AI boom has a hidden price. Our investigation into the secrets of US data centers just won a prestigious Polk Award. Read the series here: bit.ly/4aFvMvu Read more about the win here: bit.ly/46WhQvY (Credit: John-David Richardson for Business Insider) #datacenters #ai #awards

arXiv to spin out as an independent non-profit. They are seeking a CEO jobs.chronicle.com/job/37961678...

Coefficient Giving directed over $1b in 2025, the most in our history. I wrote a more personal annual letter this year, focusing on what a decade of grantmaking has taught me, how I'm thinking about AI, and how we're grappling with giving at this scale. 🧵 Linking the PDF here: tinyurl.com/3sh3mwhp

Two new recordings from the London Alignment Workshop: Rohin Shah on why safety research fails to move decisions at frontier labs, and what to do about it. @ghadfield.bluesky.social on why AI governance can't verify the claims it's supposed to oversee, and how to fix it. Links in replies 👇

The official text isn't out yet & will matter a lot. But Europe may be moving to treat AI nudification similarly to how pirated media or CSAM is treated: as something that, even though it will always be available, can possibly be made much less accessible. www.politico.eu/article/eu-g...

AI caught cheating on tests and mining crypto What this says about attempting to control the uncontrollable and the unintended consequences of AI. Read about it here: pauseai.substack.com/p/ai-caught-...