cloneofsimo drops his bearish worry that coding gains won't generalize to other domains
cloneofsimo · x · 2026-11-19
cloneofsimo shares a shifted outlook: he used to worry coding-model performance wouldn't generalize to other domains, which would require collecting similar-scale data everywhere and be extremely bearish for AGI progress. After recently seeing "some insane works," he no longer holds that worry — suggesting evidence of cross-domain generalization is accumulating. He doesn't name the specific works.
More from AGI Musings
- 'Forget human primacy or machine successionism — I just want an equal' — amplifiedamp · 2026-11-19
- Will Brown: principal-agent alignment is the easy path to avoiding AI power concentration — willcb · 2026-11-19
- Consciousness Researchers Clash Over Calling a Mainstream Position 'Ridiculous' — dioscuri · 2026-11-19
- Software Legend Grady Booch: LLMs Are a Dead End for Sentient Systems — Grady_Booch · 2026-11-19
- Formalizing Math Strips Out Intelligence, Argues Tech Commentator — LucaAmb · 2026-11-19
- OpenAI pretraining researcher: 'no confidence' humans keep better research taste than models for long — ChrisGPT · 2026-11-19