Ben Thompson argues AI's "pacing" debate is really about business overhangs, as Axios unpacks recursive self-improvement fears and OpenAI stands up a math advisory group. Plus Anthropic says Claude now leads a quarter of its own R&D, Claude Code Projects goes always-on, Google's CC expands to families, and a wildcard: a muonium beam built to test Einstein's gravity.

Ben Thompson, Stratechery · Sep 21

Examines whether AI labs' push to pace frontier model progress reflects business overhangs rather than purely safety concerns.

Axios · Sep 22

Explains recursive self-improvement and why more automated AI model training is edging labs toward a widely feared threshold.

Aditya Mehta, TechCrunch · Sep 21

OpenAI created a mathematician advisory group at Princeton's IAS after its AI system solved dozens of open math problems.

Marcus Schuler, Implicator.ai · Sep 18

Anthropic's new R&D Automation Index shows Claude now leads about a quarter of its own AI research.

Carl Franzen, VentureBeat · Sep 17

Claude Code's redesigned Projects feature coordinates parallel cloud agent threads that keep working on code after developers log off.

Tom Shane, Google Labs · Sep 17

Google expanded its CC assistant into a shared household agent letting up to six family members coordinate calendars and tasks.

Gergely Orosz, The Pragmatic Engineer · Sep 15

Reports on how the Codex coding agent became central to nearly all of OpenAI's internal engineering workflows.

Simon Willison · Sep 18

Argues that dismissing LLMs as uninteresting resembles a geneticist shrugging off the discoveries inside a newly opened Jurassic Park.

Astral Codex Ten · Sep 18

An anonymous ACX reader reviews Adam Kay's NHS junior-doctor memoir as both comedy and a case study in institutional strain.