LLM在法律推理中作为求解器与自动形式化工具的真实性
本文探讨了大语言模型（LLM）在法律推理中扮演求解器与自动形式化工具角色时的真实性问题。研究评估了模型在直接回答法律问题以及将自然语言法律文本自动形式化为可验证逻辑形式时的表现，揭示了LLM在这两项任务中可能存在的偏差与不忠实输出。
相关报道
A blog post discusses a mathematical identity where pentagonal numbers can be expressed in terms of triangular numbers. It highlights that while examples don't typically prove theorems, in this case the identity Pn = T(2n−1) − T(n−1) holds, showing that three examples can suffice for proving certain relationships.
John D. Cook describes how a sequence of his blog posts often follows a hidden thread, beginning with a post about the mathematical approximation exp(−x²) ≈ (1 + cos(sin(x) + x))/2, which some commenters incorrectly attributed solely to a first-order Taylor expansion.
The nth pentagonal number Pn follows the formula Pn = (3n² − n)/2 for positive integer n. For non-positive integer n, the same formula defines a generalized pentagonal number.
Partial fraction decomposition is commonly introduced in calculus as a technique for integrating rational functions by breaking P(x)/Q(x) into simpler terms. However, the post suggests that this method has applications beyond integration that are often overlooked in a typical calculus class.