AI RESEARCH

Fully Automated Identification of Lexical Alignment and Preference-Stage Shifts in Large Language Models

arXiv CS.CL

ArXi:2606.03165v1 Announce Type: new The language used by digital chat assistants such as ChatGPT can diverge from human expectations (misalignment). Research, mostly on Scientific English, has described both what divergences occur and, to some extent, why, linking them to the