Jump Start or False Start? A Theoretical and Empirical Evaluation of LLM-initialized Bandits
arXiv:2604.02527v1 Announce Type: new
Abstract: The recent advancement of Large Language Models (LLMs) offers new opportunities to generate user preference data to warm-start bandits. Recent studies on contextual bandits with LLM initialization (CBLI)…