POPI: Personalizing LLMs via Optimized Natural Language Preference Inference

ArXi:2510.17881v3 Announce Type: replace-cross Large language models (LLMs) are typically aligned with population-level preferences, despite substantial variation across individual users. We