The paper “Re-Centering Humans in LLM Personalization” studies how personalization systems perform when evaluated on real user conversations rather than synthetic data. The authors collected human conversations and judgments across attribute extraction, relevance matching, and personalized response generation, highlighting gaps in current LLM personalization methods.