An AI reviewer may already beat most NeurIPS reviews, says VitalyFM
vitalyFM · x · 2026-07-25
VitalyFM argues that most NeurIPS reviews are now worse than what a frontier model could produce in 15 minutes.
He adds that this judgment comes from comparing human reviews with LLM-generated ones on papers he already understood, and on public submissions only, using an LLM-based workflow in a confidential setting.
More from AGI Musings
- Surya Ganguli says physics may be a better route to understanding AI than theorem-proof theory — SuryaGanguli · 2026-07-25
- Open-weight models are how the U.S. wins on AI, says rsalakhu — rsalakhu · 2026-07-25
- Gary Marcus says China has mostly caught up in AI, especially on price — bookwormengr · 2026-07-25
- Third-party labs are needed to verify frontier AI safety claims — teortaxesTex · 2026-07-25
- Miles Brundage warns that the AI industry has a growing over-trust problem — Miles_Brundage · 2026-07-25
- MIT Sloan says AI can expand worker capability, not just automate jobs — Exp_Mark · 2026-07-25