RESEARCHInvestigateNOW
Anchoring Bias in LLM-as-a-Judge Systems: Prior Scores Compromise Evaluation Independence
arXiv cs.CL — Computation and Language
Factual evidence
What the source reports
Research demonstrates LLM-as-a-Judge evaluation independence is compromised by anchoring bias when prompts contain prior scores or history.
Open source