Author of 'Emergent Abilities a Mirage' says results valid but obvious to labs
RylanSchaeffer · x · 2026-10-20
Schaeffer reflects on his NeurIPS award-winning paper 'Are Emergent Abilities a Mirage' (with Brando and Koyejo), which grew out of a class by Percy Liang, Tatsu Hashimoto and Chris Re — whom he regrets not crediting as authors. He has mixed feelings: the results are valid but, in his view, obvious and already known inside AI labs, yet the paper accelerated his career in ways he feels he didn't deserve. The quoted tweet also covers 'Double Descent Demystified', which began as a Stanford AI Alignment tutorial reducing double descent to three factors expressible with introductory linear algebra — a lesson in simplifying phenomena to their most basic pieces.
More from AGI Musings
- Preprint maps 280 conserved fibrosis enhancers via CRISPRi CROP-seq across 549,767 cells — anshulkundaje · 2026-10-20
- Viral jab: Anthropic's hypothetical $200B public benefit fund would prioritize stopping Anthropic — birchlse · 2026-10-20
- Jason Wei: AI won't automate radiologists — elastic demand beats displacement — _jasonwei · 2026-10-20
- Language is lossy compression: a long thread argues Neuralink removes the straw entirely — r0ck3t23 · 2026-10-20
- Study: Generative AI boosts creativity only for high-metacognition users, zero gain for others — korymath · 2026-10-20
- "There are no robot foundation models, just foundation models" — tkipf · 2026-10-20