大语言模型的因果理解：不确定性的作用
Causal Understanding by LLMs: The Role of Uncertainty
AI总结：
本研究通过不确定性评估发现，LLM在因果分类中的失败源于缺乏结构化因果表征，而非预训练暴露不足，表现为随机猜测和严重校准错误。
摘要：
近期论文表明，大语言模型（LLMs）在因果关系分类中达到接近随机的准确率，这引发了关于此类失败源于预训练暴露不足还是更深层的表征差距的疑问。我们在基于不确定性的评估下对此进行研究，测试预训练期间对因果示例的暴露是否能提升因果理解能力。我们使用超过18K条PubMed句子——一半来自The Pile语料库，一半来自2024年之后——在七个模型（Pythia-1.4B/7B/12B、GPT-J-6B、Dolly-7B/12B、Qwen-7B）上进行测试。我们通过以下方式分析模型行为：（i）因果分类，即模型识别文本中的因果关系；（ii）逐字记忆探测，即评估模型是否偏好先前见过的因果陈述而非其改写版本。模型执行四分类（直接/条件/相关/无关系），并在原始句子及其生成的改写版本之间进行选择。结果显示，在已见和未见句子上准确率几乎相同（p > 0.05），没有记忆偏差（原始选择率为24.8%），且可能选项上的输出分布几乎平坦，熵值接近最大值（1.35/1.39），证实了随机猜测。指令微调模型表现出严重的校准错误（Qwen：置信度>95%，准确率32.8%，ECE=0.49）。条件关系引发最高熵（相比直接关系增加11%）。这些发现表明，因果理解失败源于缺乏结构化的因果表征，而非预训练期间因果示例暴露不足。
英文摘要：
Recent papers show LLMs achieve near-random accuracy in causal relation classification, raising questions about whether such failures arise from limited pretraining exposure or deeper representational gaps. We investigate this under uncertainty-based evaluation, testing whether pretraining exposure to causal examples improves causal understanding >18K PubMed sentences -- half from The Pile corpus, half post-2024 -- across seven models (Pythia-1.4B/7B/12B, GPT-J-6B, Dolly-7B/12B, Qwen-7B). We analyze model behavior through: (i) causal classification, where the model identifies causal relationships in text, and (ii) verbatim memorization probing, where we assess whether the model prefers previously seen causal statements over their paraphrases. Models perform four-way classification (direct/conditional/correlational/no-relationship) and select between originals and their generated paraphrases. Results show almost identical accuracy on seen/unseen sentences (p > 0.05), no memorization bias (24.8% original selection), and output distribution over the possible options is almost flat, with entropic values near the maximum (1.35/1.39), confirming random guessing. Instruction-tuned models show severe miscalibration (Qwen: > 95% confidence, 32.8% accuracy, ECE=0.49). Conditional relations induce highest entropy (+11% vs. direct). These findings suggest that failures in causal understanding arise from the lack of structured causal representation, rather than insufficient exposure to causal examples during pretraining.
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。
