A new framework called CLIPR enhances large language models' ability to infer latent user preferences from minimal interactions, improving decision-making alignment. This advancement could significantly reduce inference costs and broaden the practical applicability of LLMs in various contexts.