原帖引用 · AI 翻译
标题:“法官并不公正:医疗大语言模型评估中的自我偏好”
斯坦福大学的一篇新论文发现,接受测试的 8 个大语言模型都把自己的医疗回答排得比其他模型给同一回答的排名更高,即使回答是匿名的也是如此。
医疗 AI 基准测试常用大语言模型为其他大语言模型的回答评分。如果评判模型偏爱自己的输出,基准测试结果就会向该模型倾斜。
研究人员让来自 OpenAI、Anthropic、Google 和 Qwen 的 8 个模型回答 620 个真实的临床医生问题,然后对全部 8 个匿名回答进行排名。他们比较了每个评判模型对自己回答的排名与其他 7 个评判模型对同一回答的排名。
每个评判模型都偏爱自己的回答,平均高出 1.22 个名次;GPT-5.6 Sol 的差距接近 3 个名次。没有评分细则时,这种偏差依然存在;公布模型名称后,偏差仅缩小了 0.094 个名次。
如果使用大语言模型担任评判者,应采用由多个模型家族组成的评判小组,并分别报告各评判模型的结果,而不是只报告 1 个汇总分数。
原文
New Stanford paper finds that all 8 LLMs tested rank their own medical answers higher than other models rank them, even when answers are anonymous. Medical AI benchmarks often use LLMs to grade other LLMs' answers. If a judge favors its own outputs, the benchmark tilts toward that model. They had 8 models from OpenAI, Anthropic, Google and Qwen answer 620 real clinician questions, then rank all 8 anonymous answers. They compared where each judge placed its own answer with where the other 7 judges placed that same answer. Every judge favored its own answer, by 1.22 rank positions on average and nearly 3 for GPT-5.6 Sol. The bias held without a rubric, and revealing model names cut it by only 0.094 positions. If you use LLM judges, use a panel from several model families and report results per judge instead of 1 pooled score.