Malo Bourgon
@malo
CEO at Machine Intelligence Research Institute (MIRI, @intelligence.org)
A common question: “when should humanity hit the brakes?” @pbarnett.bsky.social’s recent paper explains pathways by which governments could lose their ability to restrain AI. We should hit the brakes while they still work! intelligence.org/wp-content/uploads/The-Closing-Window-Peter-Barnett.pdf
best paper winner from Naci Cankaya about inference verification: Model activations are bit-exact reproducible when enough information is reported, making verification easier. arxiv.org/abs/2606.00279
Recent work from Robi Rahman with Sabiha Tajdari on how to verify what chips are doing based on telemetry data. arxiv.org/abs/2606.19262
A menu of technical interventions for stopping dangerous AI development and deployment, and discussion of the role these interventions can play in various plans, by @pbarnett.bsky.social, @aaronscher.bsky.social, David Abecassis. arxiv.org/abs/2507.09801
The first comprehensive survey I’m aware of on relevant verification mechanisms (though certainly somewhat outdated given its 2024 publication) by @aaronscher.bsky.social and Lisa Thiergart. techgov.intelligence.org/research/mechanisms-to-verify-international-agreements-about-ai-development
Philosophers have spent centuries asking whether all sufficiently rational minds would converge on the same truths. At least when it comes to video games, the answer seems to be a resounding yes.
You don't have to take my word for it. LLMs are dumb in a bunch of ways, but I think this is a powerful and convincing consensus on this question.
@hankgreen.bsky.social rarely does interviews or 30+ min long videos. His latest video, an hour+ long interview with Nate Soares about “If Anyone Builds It, Everyone Dies,” is a banger. My new favorite! www.youtube.com/watch?v=5CKu...
I really quite like the stories/parables at the start of most chapters, and lots of folks who have read advance copies, from policymakers to folks in the media, have told us they really liked them too. E.g., from someone in the media:
We've been getting some pretty awesome blurbs for Eliezer and Nate's forthcoming book: If Anyone Builds It, Everyone Dies More details here: www.lesswrong.com/posts/khmpWJ... One of my favorite reactions, from someone who works on AI policy in DC:
MIRI's (@intelligence.org) Technical Governance Team submitted a comment on the AI Action Plan. Great work by David Abecassis, @pbarnett.bsky.social, and @aaronscher.bsky.social Check it out here: techgov.intelligence.org/research/res...
Pretty happy with my performance in #F1Fantasy this season. My best team finished 117th out of 2.5+ million teams! Even more impressed with my wife, who finished 79th (and at one point was in the top 50 🤯). F1 Fantasy power couple 💪.
PSA: Bluesky has an experimental “Show replies in a threaded[/nested] view” setting! It’s located in: Settings (⚙️) → Content and media → Thread preferences (You’re welcome 😉)