用首个生成词的置信度检测幻觉，效率远超传统多轮采样方法。
The First Token Knows: Single-Decode Confidence for Hallucination Detection
- 仅通过一次贪婪解码的首个有效词置信度判断幻觉
- 在多个模型和数据集上达到0.820的AUROC，优于现有方法
- 适合需要快速、低成本幻觉检测的场景
自一致性通过多次采样并测量答案一致来检测幻觉，但需重复解码且对词汇变化敏感。语义自一致性通过自然语言推理聚类采样答案含义，但增加采样成本和外部推理开销。我们发现，单次贪婪解码中首个内容词的置信度（phi_first），基于前K个logits的归一化熵计算，其表现与语义自一致性相当或略优。在三个7-8B指令微调模型和两个基准测试上，phi_first平均AUROC达0.820，高于语义一致性（0.793）和表面形式自一致性（0.791）。子集测试表明，phi_first与语义一致性中度至强相关，两者结合仅带来小幅性能提升。结果表明，多样本一致性的大部分不确定性信息已包含在初始词分布中。因此建议将phi_first作为默认低代价基线，再决定是否启用采样式不确定性估计。
原文摘要 · Abstract (English)
Self-consistency detects hallucinations by generating multiple sampled answers to a question and measuring agreement, but this requires repeated decoding and can be sensitive to lexical variation. Semantic self-consistency improves this by clustering sampled answers by meaning using natural language inference, but it adds both sampling cost and external inference overhead. We show that first-token confidence, phi_first, computed from the normalized entropy of the top-K logits at the first content-bearing answer token of a single greedy decode, matches or modestly exceeds semantic self-consistency on closed-book short-answer factual question answering. Across three 7-8B instruction-tuned models and two benchmarks, phi_first achieves a mean AUROC of 0.820, compared with 0.793 for semantic agreement and 0.791 for standard surface-form self-consistency. A subsumption test shows that phi_first is moderately to strongly correlated with semantic agreement, and combining the two signals yields only a small AUROC improvement over phi_first alone. These results suggest that much of the uncertainty information captured by multi-sample agreement is already available in the model's initial token distribution. We argue that phi_first should be reported as a default low-cost baseline before invoking sampling-based uncertainty estimation.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。