What Evidence Do Language Models Find Convincing? (Wan et al. 2024)

反直觉奠基：LLM 重相关性，几乎无视人类看重的引用/中立语气等风格信号（ACL'24 Main）