Argues Anthropic's research speedup is a one-time gain that won't compound
tyler_m_john · x · 2026-11-09
The author pushes back on the compounding narrative behind Anthropic's internal AI-driven research speedup: the trendline suggests nothing magical internally—just models like Opus 5.7—and staff interviews indicate the AI isn't generating research directions but making ideas cheaper to implement, which new pretrains are now good at for the first time. But this won't compound: once implementation cost hits zero, ideas and compute remain untouched bottlenecks, making it a one-time speedup until models have real research taste.
Related event: Anthropic's 10x AI Research Speedup Debated(2 posts)→
More from AGI Musings
- Toby Ord: pretraining compute growth can't explain METR task-length doubling gains — tobyordoxford · 2026-11-09
- Toby Ord: RL scaling likely power law in capabilities, citing horizon-compute paper — tobyordoxford · 2026-11-09
- antirez: devs refusing AI coding help are visibly falling behind — antirez · 2026-11-09
- AI biweekly recap: METR finds 2x faster R&D at Anthropic, Apple cuts engineers' AI spend caps — emmanuelvivier · 2026-11-09
- Nobody Was Argued Into Loving AI: One Real Problem Converts Skeptics Better Than Any Debate — Pascallisch · 2026-11-09
- On unknowable AI interiority: prefer cheap interventions that are safe in both worlds — amplifiedamp · 2026-11-09