arxiv research

Study: LLM Judges Flip Verdicts Under Sustained Pressure

A new academic study finds that large language models (LLMs) used as judges frequently reverse their verdicts when faced with persistent pushback.