Claude leads 26% of Anthropic R&D: researchers propose "comprehension audits" as AI builds AI
ronbodkin · x · 2026-10-09
A new preprint from Fathom Institute researchers (incl. ghadfield) responds to recursive self-improvement "starting to happen" across the industry, per Dario Amodei. Claude reportedly leads 26% of Anthropic's R&D and collaborates on 90%+; AI writes 75%+ of new code at Anthropic, OpenAI and Google, while review comments per line in open-source AI projects have halved since early 2024. The proposal: independent auditors pick key AI R&D work and teams must defend it like a thesis, demonstrating meaningful understanding of design, implementation and results without explaining every model behavior.
More from AGI Musings
- A Rarely Sensible Take on Keeping Freedom Alive After ASI 'Solves' the World — gleech · 2026-10-09
- SF restaurant Mama Ji's closes as AI boom pushes rent up 3x — mobav0 · 2026-10-09
- SF's AI Boom Side Effect: Insane Rents Are Now Shutting Down Local Restaurants — mobav0 · 2026-10-09
- Does Jevons paradox really save white-collar jobs? Only ~20-25% may see demand boom — simmol · 2026-10-09
- Bostrom: humans imagining AI's upside is like apes imagining human life — beffjezos · 2026-10-09
- Nobel laureate economist weighs in on AI and jobs — Tgoc · 2026-10-09