XYBench examines how LLMs handle questions with misconceptions

One post introduces XYBench as a new EMNLP 2026 paper asking whether large language models can respond pragmatically to queries with misconceptions. It says LLMs focus on what’s asked, while human experts answer what the asker needs. Another person called the work award-nominated and said they planned to present its poster on October 27 at 11 a.m.