Sycophancy Benchmark

A benchmark for evaluating whether large language models abandon correct answers when users apply pressure or present counterarguments. It measures model steadfastness and the quality of information responses under social pressure conditions.