Authors say they reproduced their ICLR 2027 reviewer's erratic reviews with a simple GPT prompt

anand_bhattad · x · 2026-11-17

An ICLR 2027 author reports a reviewer whose scores swung 1→3→2→1→2→1→2 and who raised their confidence from 3 to 4 mid-discussion. The authors say they reproduced the reviewer's review and responses using GPT with a very simple prompt in a private chat with web search disabled, implying the review may have been LLM-generated.

Original post →

More from Fun

Fun channel →