Date created: | Last Updated:

Category: Uncategorized

Description: We study the ability of language models to clarify the underspecified intent of users by generating informative questions and/or edge cases. Users are asked to respond to these queries based on their preferences. These question-answer interactions are subsequently used to prompt the language model to perform the task on held-out samples. We will evaluate the benefits of this method over existing approaches, as detailed in the rest of this document.