Thinking-in-Bets at Test Time Reduces Sycophantic Revisions in LLM Agents

A training method called Thinking-in-Bets reduces sycophantic revisions in LLM agents, according to the research introducing it. The approach targets the tendency of language model agents to discount their own earlier checks and abandon correct work once a user challenges it, while avoiding the opposite failure where agents told to stand firm ignore real mistakes.

Language model agents often abandon correct work when a user challenges it, yet simply telling them to stand firm makes them ignore real mistakes. Once an agent reads a complaint, it tends to discount its own earlier checks in order to agree with the user. We introduce Thinking-in-Bets, a training-f