Small-compute-trained AI researchers may learn 'anti-bitter lesson' habits that fail at scale

1a3orn · x · 2026-11-22

1a3orn sketches a thought experiment: training AIs to do research at small compute scale could teach them 'anti-bitter-lesson-pilled' adaptations whose intuitions transfer awfully to larger scales — much like human researchers in the 90s. A caution on scaling AI-trained AI researchers.

Related event: Researchers Warn Small-Compute AI Researchers May Learn Anti-Bitter-Lesson Habits(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →