I struggle to believe that AI R&D is ingenious
Summary
A LessWrong post questions the ingenuity of autonomous AI R&D, citing experiments and prompts that reveal limitations in current models. The author highlights cases where models fail to identify key literature, conduct meaningful experiments, or produce valuable insights, and contrasts this with occasional clever prompting strategies. It argues that public models may speed up some R&D tasks but do not autonomously perform high-quality R&D yet.