Do Large Language Models have Shared Weaknesses in Medical Question Answering?

📄 arXiv: 2310.07225v3 📥 PDF

作者: Andrew M. Bean, Karolina Korgul, Felix Krones, Robert McCraith, Adam Mahdi

分类: cs.CL

发布日期: 2023-10-11 (更新: 2024-10-11)

备注: 8 pages, 10 figures. To appear in NeurIPS 2024 Advancements in Medical Foundation Models Workshop

💡 一句话要点

评估大型语言模型在医学问答中的共性弱点

🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 大型语言模型 医学问答 基准测试 模型性能 人类表现 问题难度 准确率分析

📋 核心要点

  1. 现有大型语言模型在医学问答中表现不稳定，影响其在实际应用中的安全性和可靠性。
  2. 本文通过基准测试多种顶尖LLMs，旨在识别模型间的共性优缺点，以便为未来模型设计提供指导。
  3. 实验结果显示，模型准确率与人类表现存在一定相关性，且问题长度对准确率有负面影响，最优模型的准确率达到84%。

📝 摘要（中文）

大型语言模型（LLMs）在医学基准测试中取得了快速进展，但其不可靠性仍然是安全应用的持续挑战。为了解决这一问题，本文对多种顶尖LLMs进行了基准测试，识别出模型间的共性优缺点。研究测试了16个知名LLMs在874个新收集的波兰医学执照考试问题上的表现，分析了模型的准确率与人类表现的相关性，并探讨了问题难度、长度等因素对模型表现的影响。结果表明，模型的准确率与信心呈正相关，而与问题长度呈负相关。最优模型GPT-4o Turbo的准确率为84%。

🔬 方法详解

问题定义：本文旨在解决大型语言模型在医学问答中表现不一致的问题，现有方法缺乏对模型间共性优缺点的系统性分析。

核心思路：通过对16个知名LLMs在874个医学考试问题上的表现进行基准测试，识别出模型间的共性表现和影响因素，以便为未来的模型设计提供参考。

技术框架：研究首先收集了874个医学问题，然后对每个模型进行评分，分析其准确率和概率分布，最后比较与人类表现及其他模型的相关性。

关键创新：本研究的创新点在于系统性地比较多种LLMs的表现，识别出它们在回答正确问题上的相似性，以及与人类考生的相似性，从而为理解模型的共性弱点提供了新的视角。

关键设计：在实验中，使用了准确率和概率分布作为主要评估指标，分析了问题的难度、长度等因素对模型表现的影响，发现模型的准确率与信心呈正相关，而与问题长度呈负相关。

📊 实验亮点

实验结果显示，顶级模型GPT-4o Turbo在874个医学问题上的准确率达到84%，而Claude Opus、Gemini 1.5 Pro和Llama 3/3.1的准确率在74%到79%之间。模型表现与人类表现的相关性较低，且问题长度对准确率有显著负面影响，提供了对模型设计的重要见解。

🎯 应用场景

该研究的潜在应用领域包括医学教育、临床决策支持和医疗信息检索等。通过识别大型语言模型在医学问答中的共性弱点，研究为未来模型的改进和安全应用提供了重要参考，可能推动更可靠的医疗AI系统的开发。

📄 摘要（原文）

Large language models (LLMs) have made rapid improvement on medical benchmarks, but their unreliability remains a persistent challenge for safe real-world uses. To design for the use LLMs as a category, rather than for specific models, requires developing an understanding of shared strengths and weaknesses which appear across models. To address this challenge, we benchmark a range of top LLMs and identify consistent patterns across models. We test $16$ well-known LLMs on $874$ newly collected questions from Polish medical licensing exams. For each question, we score each model on the top-1 accuracy and the distribution of probabilities assigned. We then compare these results with factors such as question difficulty for humans, question length, and the scores of the other models. LLM accuracies were positively correlated pairwise ($0.39$ to $0.58$). Model performance was also correlated with human performance ($0.09$ to $0.13$), but negatively correlated to the difference between the question-level accuracy of top-scoring and bottom-scoring humans ($-0.09$ to $-0.14$). The top output probability and question length were positive and negative predictors of accuracy respectively (p$< 0.05$). The top scoring LLM, GPT-4o Turbo, scored $84\%$, with Claude Opus, Gemini 1.5 Pro and Llama 3/3.1 between $74\%$ and $79\%$. We found evidence of similarities between models in which questions they answer correctly, as well as similarities with human test takers. Larger models typically performed better, but differences in training, architecture, and data were also highly impactful. Model accuracy was positively correlated with confidence, but negatively correlated with question length. We find similar results with older models, and argue that these patterns are likely to persist across future models using similar training methods.