ICML 2027 chair reveals SOTA LLM desk-rejects ~90% of submissions
ZeeshanZiaML · x · 2026-09-21
Andrew G. Wilson, Program Chair for ICML 2027, publicly asked for suggestions on handling exploding submission counts, AI slop, reviewing burdens, and bureaucracy.
One ML researcher replied that the prospective desk is reportedly using a SOTA LLM to desk-reject about 90% of submissions, and recommended tuning the LLM prompt on a small PC-level test set before deployment. It's a rare glimpse of top venues embedding LLMs directly into the triage pipeline to fight AI-generated slop.
More from Research
- ICML 2027 chair asks for ideas; scholar proposes 5 fixes for submission explosion — mircomusolesi · 2026-09-22
- Human video emerges as robotics pretraining substrate: effective experience = hours × information per hour — rohanpaul_ai · 2026-09-22
- Maxinsights report: Physical AI data scaling laws — after 1M hours, it's experience density — rohanpaul_ai · 2026-09-22
- 4B Qwen tuned with SFT+RL produces 81% faster Postgres query plans for $1,200 — JeremyCMorgan · 2026-09-22
- OpenAI's internal math model said to have solved 100+ long-standing open problems — inductionheads · 2026-09-22
- Jayce: framework-free prototype memory lets local LLMs learn facts 1.6x-4x faster than backprop — kavanutz · 2026-09-22