让大模型推理更可信，能自动学习判断真假并保留更多正确结论。
Differentiable Conformal Training for LLM Reasoning Factuality
- 用可微分方法优化大模型的真假判断逻辑，实现端到端训练。
- 在两个基准数据集上，保留率提升141%且保持低于10%的幻觉率。
- 适合需要高可靠性推理的场景，如医疗、金融等关键领域。
大语言模型常出现幻觉，限制其在关键应用中的可靠性。置信预测（CP）通过校准保留数据上的误差率，提供统计有效的置信保证。近期工作将CP扩展至大模型事实性，以过滤高风险陈述，确保幻觉率低于用户指定水平（如10%）。以往方法独立处理每个陈述，而连贯事实性通过依赖图表示输出，联合验证陈述及其逻辑祖先。其主要局限是不可微分，需手工设计评分器，在高可靠性水平下几乎剔除60%的真实陈述。我们提出可微分连贯事实性（DCF），一种完全可微分的松弛方法，可在保证恢复原始算法有效性的同时学习改进评分器。在两个基准推理数据集上的实验表明，DCF在保持可靠性保证的前提下，最多实现141%的陈述保留率提升，标志着迈向可靠置信大模型系统的重要一步。
原文摘要 · Abstract (English)
Large Language Models (LLMs) frequently hallucinate, limiting their reliability in critical applications. Conformal Prediction (CP) addresses this by calibrating error rates on held-out data to provide statistically valid confidence guarantees. Recent work extends CP to LLM factuality to filter out risky claims, ensuring that hallucination rates remain below a user-specified level (e.g., 10%). While prior methods treat claims independently, Coherent Factuality extends to multi-step reasoning by representing outputs as dependency graphs and jointly validating claims with their logical ancestors. A key limitation is that Coherent Factuality is not differentiable, requiring hand-crafted scorers that at high reliability levels remove nearly 60% of true claims. We introduce Differentiable Coherent Factuality (DCF), a fully differentiable relaxation that enables learning improved scorers while provably recovering the original algorithm's guarantees. Experiments on two benchmark reasoning datasets demonstrate DCF achieves up to 141% improvement in claim retention while maintaining reliability guarantees, representing a significant step towards reliable conformal LLM systems.
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。