everyone keeps saying frontier labs need to present a more compelling vision for the future. a world where all of humanity shares in the benefits. but nobody ever says how. i'd like to propose we start with a pizza party

Chekhov's superintelligence— the dominant AI narrative largely presupposes two things: 1) it might kill us 2) it is going to be built. A sort of narrative inevitability before its technical inevitability has been properly established. Practically, even if progress stopped today—and it isn't, hello, Opus 5.5—what already exists has changed us and continues to diffuse unevenly through society and culture; this alone is enough to drastically reshape our way of life. Psychically, we've already accepted that this future is something we have to absorb and make sense of; this is most of what "AI has happened" actually means. What surprises me is how deep that acceptance goes—even people who want AI development curtailed expect it to continue. The public broadly—most of whom are unfamiliar with AI—seem to accept it too. Perhaps we are more technological people than I supposed, well instructed by iPhone and TikTok. Acceptance is not approval, but to me it does seem notable that inevitability as a starting place has always already been conceded—in sharp contrast to something like climate change, where decades were spent arguing about the particularity of whether/how it was happening. Climate denial was largely supported by political coalitions and industries the premise threatened. Here, the industry endorses and more importantly benefits from inevitability, so no one powerful argues against it. AI arrived culturally primed and quickly diffused in ways that made it hard to deny its salience for consumers. Prominent figures do trouble the prognosis of near-term powerful AI (LeCun, Marcus, Bender) but these skeptics have almost no narrative power. Dissent from the premise exists, but it doesn't meaningfully shape the discursive landscape—if anything it serves as a foil for the inevitability. The dissent that does seem to have narrative power, precipitated by the recent Anthropic whistleblower, doesn't dispute that powerful AI is coming; it warns that it's dangerous and asks us to slow down. Chekhov's gun is a fun thing to write and now I have to find a way to fit it in. It's a narrative tool—advice to playwrights, not some sort of world law. It's the audience's expectation that makes the shot feel inevitable. Every warning draws our attention back to the gun. But here the playwright believes in the gun even more than the spectator.

I've added the reading list for our MATS stream (co-led with @LRudL_) to my website. It's a good starting place for AI strategy & governance work, especially if you're focused on distributing power/preventing power concentration. A 🧵 on some of my favorite recommendations:

claude opus velocity (version gain / 100 days) 4.1 █ 0.13 4.5 ████ 0.36 4.6 █ 0.14 4.7 █ 0.14 4.8 ██ 0.24 5 ████ 0.35 5.5 ████████ 0.83 4 → 4.8 took a year. 4.8 → 5.5 took 4 months.