MedCalc-Bench 并未衡量你所想的:一项基准审计与开放书考试的案例
MedCalc-Bench Doesn't Measure What You Think: A Benchmark Audit and the Case for Open-Book Evaluation
AI总结:
MedCalc-Bench的基准审计揭示其主要衡量公式记忆和算术精度,而非临床推理,通过开放书提示显著提升模型性能。
摘要:
MedCalc-Bench 是一个广泛用于评估大语言模型在临床计算器任务上性能的基准,其最先进的直接提示得分在验证集上停滞在约35%(HELM MedHELM排行榜),而最好的已发表方法-强化学习(RL)在可验证奖励下达到74%。我们提出了三个贡献,挑战了该基准当前的框架。首先,我们对基准的计算器实现进行了系统审计,发现了并修复了超过20个错误,从关键公式不准确到NeurIPS发表数据集中运行时的bug。其次,我们展示了简单的干预-在推理时为模型提供计算器规范("开放书"提示)-将准确性从约52%提升到GLM-4.6V和GLM-4.7上的81-85%,超越了所有已发表的结果,包括RL训练的系统,无需任何微调。第三,我们使用GPT-5.2-Thinking建立了95-97%的上限,残余误差主要归因于真实情况问题和数据集的歧义。我们的发现表明,MedCalc-Bench主要衡量的是公式记忆和算术精度,而非临床推理,并应被重新定位为工具使用评估。
英文摘要:
MedCalc-Bench is a widely used benchmark for evaluating LLM performance on clinical calculator tasks, with state-of-the-art direct prompting scores plateauing around 35% on the Verified split (HELM MedHELM leaderboard) and the best published approach-RL with verifiable rewards-reaching 74%. We present three contributions that challenge the benchmark's current framing. First, we conduct a systematic audit of the benchmark's calculator implementations, identifying and fixing over 20 errors ranging from critical formula inaccuracies to runtime bugs in a NeurIPS-published dataset. Second, we show that a simple intervention-providing the model with the calculator specification at inference time ("open-book" prompting)-raises accuracy from ~52% to 81-85% on GLM-4.6V and GLM-4.7, surpassing all published results including RL-trained systems, without any fine-tuning. Third, we establish an upper bound of 95-97% using GPT-5.2-Thinking, with residual errors attributable primarily to ground-truth issues and dataset ambiguities. Our findings suggest that MedCalc-Bench predominantly measures formula memorization and arithmetic precision rather than clinical reasoning, and would be better framed as a tool-use evaluation.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。