AnalysisAI ModelsNovember 6, 2026

LessWrong post questions whether autonomous AI R&D is genuinely ingenious

Read original source →lesswrong.com

Author monitored side projects and judged an Apart hackathon where eight AI-written papers scored 60-95% on Pangram but were often near-identical and methodologically incomplete. Exception: prompting Astra repeatedly improved SOTA on "listing all legal Scrabble moves" by 10x.

1 source

More stories today

Open the live feed