Tag
This paper introduces an online contextual Pandora's Box model for adaptively querying and selecting LLM APIs, proposing a learning approach that combines GMM estimation with UCB-style confidence bounds and proving dimension-dependent regret bounds.