Anthropic Researcher: New Model's Training Was Mostly One Person Working <2 Weeks
clarejtbirch · x · 2026-11-21
Anthropic's cHHillee reveals that the majority of a newly announced model's training was done by <10% of one researcher (plus his agents) over a couple of weeks, building on foundations from the rest of the team, with blog/evals/safety testing/demos/partnerships requiring heroic last-mile work. His takeaway: the company's "model factory" has progressed to the point where a single person can drive a major training run.
More from Companies & People
- AMD AI DevDay Paris on Dec 12: vLLM Maintainer to Talk Decode Optimization, Free ROCm Cert — AccBalanced · 2026-11-21
- Trapit Bansal has left Meta, per report — mrdrozdov · 2026-11-21
- Musk Probes Cursor Over Open-Source Models, X Jokes About Hiding Chinese Models — TinfoilTricorn · 2026-11-21
- Empirical Machines is hiring AI-enabled chemists and material scientists — shyamalanadkat · 2026-11-21
- Investor: Gemini turnaround shows OpenAI could retake the lead in days — bindureddy · 2026-11-21
- Podcast: AI bubble, lab IPO timing, and San Francisco's crisis of purpose — hardimanjames · 2026-11-21