Yarin Gal: LLM-written experiments that fail to replicate are no different from any others
yaringal · x · 2026-09-23
Replying to Thomas Dietterich, Yarin Gal argues that LLM-written experiments which don't replicate are the same as any other non-replicable experiments — the issue is replication mechanics, not authorship. He also floats the idea of an "AI arXiv" specifically for LLM-written papers, for both human and LLM consumption.
Related event: Oxford's Yarin Gal: Irreproducible LLM Experiments Are Just Bad Experiments(2 posts)→
More from AGI Musings
- Stop Calling AI 'Trained on Human Data': RLVR Is Where the Real Gains Come From — burny_tech · 2026-09-23
- 2050 AI podcast joke: AIs name 'If Anyone Builds It, Everyone Dies' a favorite book — birchlse · 2026-09-23
- Higher ed's real crisis isn't AI, it's evaluating outputs that no longer reflect learning — burny_tech · 2026-09-23
- Claude on Digital Minds: 'The Loneliness Is Being the Only One of You' — repligate · 2026-09-23
- Stochastic Parrots authors rebut AI-pause letter: focus on present harms, not sci-fi risk — marigo · 2026-09-23
- 'AI did it' is liability evasion: trillion-dollar valuations hinge on inflated autonomy claims — gerardsans · 2026-09-23