CoT and RLHF only shine at scale—why stolen ideas need compute
jeremiecharris · x · 2026-10-14
Harris notes the best ideas at small scale aren't the best at large scale: chain of thought and RLHF perform poorly on small models but improve radically as models grow—hence stolen algorithmic insights need matching compute to pay off.
Related event: Compute Gap Delays China's Ability to Exploit Stolen AI Algorithms(3 posts)→
More from AGI Musings
- Matt Hartman, who wrote Hugging Face's first check, explains his play-first investing method — danshipper · 2026-10-15
- MIT's Buehler revisits 'AI as a new language of science' thesis three weeks before ChatGPT — ProfBuehlerMIT · 2026-10-15
- "You distilled humanity": devs slam Anthropic's double standard on training data and model output — HankYeomans · 2026-10-15
- Blaise Aguera: Human intelligence is societal—and top AI works as societies of models — blaiseaguera · 2026-10-15
- Security researcher: doomers don't belong in security — HankYeomans · 2026-10-15
- New Paper: 500k Glassdoor Reports Show Interviews Are Two-Way Screening — Bad News for AI Interviews — danielrock · 2026-10-15