TL;DR
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
Astral Codex Ten’s weekly Open Thread 453 announced a grant program of up to $10 million from Astralis Foundation and Macroscopic Ventures for work on how countries beyond the US and China can shape AI development. The thread also links a rebuttal from Jamie Joyce, the challenger in California’s 12th District House race whom the author declined to endorse in the primary, and rounds up comments of the week on AI generalization research.
The blog Astral Codex Ten published its weekly Open Thread 453, carrying three main announcements: a grant program of up to $10 million for work on “middle powers and transformative AI,” a response from Jamie Joyce — the challenger in California’s 12th Congressional District House race — to the blog’s earlier primary endorsement decision, and new effective altruism meetup groups at three universities. The post matters to readers because ACX functions as a hub for the rationalist and effective altruism communities, and items posted there often signal where community funding, political attention, and research debate are heading.
The largest item in the thread is the grant announcement. Two organizations, Astralis Foundation and Macroscopic Ventures, will distribute up to $10 million in grants for work on what the announcement calls “middle powers and transformative AI.” The blog describes the scope as “research, training, public writing and new organizations focused on what countries beyond the US and China can do to help navigate” the AI transition. The author, who writes under the name Scott Alexander, notes uncertainty about the funders’ relationship to the broader community ecosystem, saying he thinks they are “parts of our conspiracy although I can’t keep track any more.” Interested parties can read more and apply through the linked announcement.
The thread also returns to a political controversy from earlier in the year. In his California primary ballot recommendations post this spring, the author said he was voting for the “boring incumbent” in the 12th District House race because the challenger “seemed eccentric and had no chance.” That challenger, Jamie Joyce, contacted him and asked him to change his mind — and because both candidates will face each other again in the November general election, the exchange has renewed relevance. The author declined to relitigate his May position, writing that he “may reconsider if I make a post on the November election,” but agreed to link her response, which he recommends to interested 12th District voters.
The remaining announcements are community-focused. New effective altruism meetup groups are now planned at the University of Oxford, University College Dublin, and the University of the Philippines, extending a running series of updates on college EA groups. The thread also highlights four “comments of the week,” all drawn from the blog’s recent post on the “Mysteries of AI Generalization”: Eliezer Yudkowsky on why recent generalization failures are bad news; Ezra Newman on distinctions between pretraining, midtraining, and RLVR; Elias Schmied linking Evan Hubinger on why “Hacker Opus” did not show emergent misalignment; and Stuart Armstrong on how emergent misalignment happens.
Why the $10 Million Grant Matters
The grant program is the thread’s most consequential item because it directs substantial money toward an under-examined slice of AI policy: the role of countries that are neither the United States nor China. Most AI governance debate concentrates on the two leading powers, but middle powers — through regulation, compute access, talent, and diplomacy — could shape how transformative AI is developed and deployed globally. Funding for “research, training, public writing and new organizations” suggests the program aims to build a field, not just fund isolated papers.
The political item matters at the local level. California’s 12th District is a safely Democratic seat, so the general election outcome is rarely in doubt, but ACX’s endorsement posts reach a readership that extends well beyond the district, and the exchange shows a candidate directly engaging an influential online commentator to contest his characterization of her. Whether the author revisits the race before November could shape how his readers view it.
The comments of the week, meanwhile, illustrate where technical debate in this community currently sits: on how and why AI models generalize — or fail to — and whether emergent misalignment is a real mechanism or an artifact of training methods. These threads feed directly into broader arguments about AI risk.
The Open Thread and Its Ecosystem
Open Thread 453 is part of a long-running weekly series on Astral Codexten in which readers can post on any topic and ask questions. The blog, written by Scott Alexander, grew out of the earlier Slate Star Codex and is a central node for the rationalist and effective altruism communities. As the post notes, ACX has an unofficial subreddit, a Discord server, a bulletin board, and in-person meetups around the world; most content is free, with some posts reserved for subscribers.
The ballot recommendation post referenced in the thread was published this spring ahead of California’s primary. The author endorsed the incumbent in the 12th District — the seat long associated with San Francisco — while describing the challenger as eccentric and unlikely to win. Because California’s top-two primary system sends the two leading vote-getters to the general election, both candidates will appear on the November ballot regardless.
The “Mysteries of AI Generalization” post that generated the highlighted comments is part of the blog’s regular coverage of machine learning research, where commenters including researchers and longtime AI risk commentators debate interpretation of recent results.
“I don’t want to relitigate what I said in May, but I may reconsider if I make a post on the November election. In the meantime, she asked me to link her response, which I recommend to any interested 12th District voters.”
— Scott Alexander, Astral Codex Ten
Unanswered Questions About the Grants and Endorsement
Several things remain unclear. The post does not specify the deadline, application process details, size of individual grants, or evaluation criteria for the $10 million program; readers must consult the linked announcement for specifics. The author himself expresses uncertainty about how Astralis Foundation and Macroscopic Ventures relate to the broader effective altruism funding ecosystem, describing them only tentatively.
On the political item, it is not yet known whether the author will publish a November election endorsement post, or whether he would change his position if he does. Joyce’s response is linked but the thread does not summarize her arguments, so readers must read her statement directly to evaluate them. The allocation of the grant money and the outcomes of the highlighted AI generalization debates also remain open.
What to Watch Through November
Readers can expect the next open thread in the weekly series, which will likely carry further college EA meetup updates and community announcements. The main developments to watch are whether the author publishes a ballot recommendations post for the November general election and whether it revisits the 12th District race now that Joyce has publicly responded. On the funding side, the middle-powers grant program will move into its application and selection phase; the announcement of grantees would give a concrete picture of what “middle powers and transformative AI” work looks like in practice. The technical discussion of AI generalization and emergent misalignment, seeded by the comments of the week, is likely to continue in future posts and open threads.
Key Questions
What is Open Thread 453?
It is the 453rd weekly “visible open thread” on the blog Astral Codex Ten, where readers can post on any topic. This edition included announcements about a $10 million AI grant program, a California House candidate’s response to the blog’s primary endorsement, and new university EA meetup groups.
Who is giving out the $10 million in grants?
The grants come from the Astralis Foundation and Macroscopic Ventures, for work on “middle powers and transformative AI” — specifically research, training, public writing, and new organizations focused on what countries beyond the US and China can do about AI development. The blog links to the application page for details.
In his spring primary recommendations, the author said he voted for the incumbent in California’s 12th District because Joyce, the challenger, seemed eccentric and had no chance. Joyce asked him to reconsider and to link her response ahead of the November general election, where both candidates will face off again.
That is unclear. He wrote that he does not want to relitigate his May comments but “may reconsider” if he makes a post on the November election, and he recommended Joyce’s response to interested voters.
What were the comments of the week about?
All four highlighted comments came from the blog’s “Mysteries of AI Generalization” post, covering why generalization failures concern AI risk advocates, distinctions between training stages such as pretraining and RLVR, and competing explanations for emergent misalignment in models.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
