Stanford researcher warns against using LLMs to audit scientific literature
anshulkundaje · x · 2026-11-19
Computational biologist Anshul Kundaje pushed back on the idea of using AI to audit the scientific literature. He argues LLMs lack judgment: distinguishing good science from actual error involves enormous subtlety, and today's models aren't smart enough — the likely outcome is tremendous make-work and witch hunts.\n\nHe isn't opposed to fixing errors, noting that for computational work it isn't hard to go in and determine which bugs are critical versus immaterial — implying targeted human review over blanket LLM auditing.
Related event: Researchers push back on using LLMs to audit scientific literature(5 posts)→
More from AGI Musings
- AI Velocity Ledger catalogs 41 verified feats of machine-accelerated discovery — JosephJacks_ · 2026-11-22
- Lawyer Pushback: Passing the Bar Doesn't Mean AI Can Argue in Court Yet — TinfoilTricorn · 2026-11-22
- Rebutting Cory Althoff: Superintelligence Already Happened — TinfoilTricorn · 2026-11-22
- GRH moment shows knowledge isn't just truth — it's about accessibility, says dev — AntonObukhov1 · 2026-11-22
- Yudkowsky's 'rolling cylinder' defense draws fire: 'too many sycophants around you' — teortaxesTex · 2026-11-22
- Users want frontier AI at bargain prices while complaining labs aren't profitable — VraserX · 2026-11-22