本論文では、大規模言語モデル（LLM）が法的推論において、問題解決（ソルバー）と法規の自動形式化（オートフォーマライザー）という二つの役割を果たす際の「忠実性」を評価する。具体的には、自然言語で記述された法的ルールを形式言語に変換し、その変換が元の意図を正確に保持しているか、また形式化されたルールに基づく推論結果が正しいかを検証する。実験の結果、LLMは自動形式化において高い忠実性を示す一方、複雑な法的推論タスクでは一貫性に課題があることが明らかになった。
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.