DiscoverLLM: From Executing Intents to Discovering Them

To handle ambiguous and open-ended requests, Large Language Models (LLMs) are increasingly trained to interact with users to surface intents they have not yet expressed (e.g., ask clarification questions). However, users are often ambiguous because they have not yet formed their intents: they must observe and explore outcomes to discover what they want. Simply asking "what kind of tone do you want?" fails when users themselves do not know. We introduce DiscoverLLM, a novel and generalizable framework that trains LLMs to help users form and discover their intents. Central to our approach is a novel user simulator that models cognitive state with a hierarchy of intents that progressively concretize as the model surfaces relevant options---where the degree of concretization serves as a reward signal that models can be trained to optimize. Resulting models learn to collaborate with users by adaptively diverging (i.e., explore options) when intents are unclear, and converging (i.e., refine and implement) when intents concretize. Across proposed interactive benchmarks in creative writing, technical writing, and SVG drawing, DiscoverLLM achieves over 10% higher task performance while reducing conversation length by up to 40%. In a user study with 75 human participants, DiscoverLLM improved conversation satisfaction and efficiency compared to baselines.

When you ask an AI to tackle an open-ended task like "write me a poem about an animal" or "design a logo," it usually produces a finished result on the first try. But often you don't know what you want until you see it---you might realize that you want a "less boring" poem only after reading it. Today's AI assistants assume you already know your goals from the start and just need to spell them out. When you genuinely haven't figured them out yet, the assistant asking "less boring in what way?" doesn't help, because you can't answer. We built a training method called DiscoverLLM that teaches AI to help people figure out what they want, rather than just executing requests. The key idea is a simulated user that gradually realizes more specific preferences as the AI shows relevant options, which lets us reward the AI for helping that discovery happen. Trained this way, the AI learns when to explore possibilities and when to commit and refine. Across diverse tasks, and in a study with 75 people, our AI helped people reach good results faster, and they felt it seemed to "anticipate" what they were looking for.