Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DiscoverLLM: From Executing Intents to Discovering Them

Tae Soo Kim, Yoonjoo Lee, Jaesang Yu, John Joon Young Chung, Juho Kim

TL;DR

DiscoverLLM addresses the challenge of ill-defined user intents in open-ended tasks by formalizing intent discovery as a hierarchical, co-evolving process and introducing a cognitively grounded user simulator. The framework uses a ground-truth intent tree and a per-turn reward combining discovery progress and an efficiency penalty to fine-tune LLMs toward exploring latent intents before converging on concrete outputs. Across creative writing, technical writing, and SVG drawing, DiscoverLLM improves intent discovery by about 10%, increases interactivity, and reduces conversation length, with generalization to unseen domains and a positive user study demonstrating higher satisfaction and efficiency. The work advances human-centric AI by equipping models to collaboratively explore and shape user goals, while also acknowledging safety and evaluation limitations and outlining future directions for broader validation.

Abstract

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.

DiscoverLLM: From Executing Intents to Discovering Them

TL;DR

DiscoverLLM addresses the challenge of ill-defined user intents in open-ended tasks by formalizing intent discovery as a hierarchical, co-evolving process and introducing a cognitively grounded user simulator. The framework uses a ground-truth intent tree and a per-turn reward combining discovery progress and an efficiency penalty to fine-tune LLMs toward exploring latent intents before converging on concrete outputs. Across creative writing, technical writing, and SVG drawing, DiscoverLLM improves intent discovery by about 10%, increases interactivity, and reduces conversation length, with generalization to unseen domains and a positive user study demonstrating higher satisfaction and efficiency. The work advances human-centric AI by equipping models to collaboratively explore and shape user goals, while also acknowledging safety and evaluation limitations and outlining future directions for broader validation.

Abstract

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.
Paper Structure (90 sections, 5 equations, 9 figures, 6 tables)

This paper contains 90 sections, 5 equations, 9 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Problem Formulation
  3. Intent Discovery Through Interaction
  4. Response-Driven Progressive Discovery.
  5. User Expressiveness Constraint.
  6. Objectives
  7. Underlying Assumptions
  8. DiscoverLLM: General Training Framework for Intent Discovery
  9. Operationalizing Intents as a Hierarchy
  10. Discovery States.
  11. Response Evaluation & State Transitions (\ref{['fig:user_simulator']}B)
  12. State Transitions.
  13. Constructing Intent Trees (\ref{['fig:user_simulator']}A)
  14. Reward Function
  15. Discovery Progress ($\text{R}_{\text{d}}$).
  16. ...and 75 more sections

Figures (9)

  • Figure 1: DiscoverLLM Framework: A simulated user with a latent intent hierarchy (1) interacts with a model. The user can only articulate discovered intents (2), and model responses (3) that successfully probe or satisfy undiscovered intents trigger state updates (4). The framework computes rewards based on discovery progress (5), which are used for fine-tuning of the model (6).
  • Figure 2: (A) Intent tree construction: (1) Given an artifact type and its content, we generate a specific intents list, (2) iteratively abstract them across levels, and (3) organize all resulting intents into a tree hierarchy. (B) Simulation Example: The user simulator begins ($t=0$) with only a few abstract intents discovered and provides an initial request based on these. At $t=1$, the model's response fails to probe or satisfy intents in the refinement space so no state updates occur and the user remains vague. At $t=2$, the model provides various options where one probes at an undiscovered intent, updating its state and enabling the user to articulate it.
  • Figure 3: Behavioral patterns across Qwen3-8B variants in Creative Writing. Turns are classified as divergent (D) or convergent (C) and analyzed as trigrams. Base is almost entirely convergent (91% CCC), SFT overfits toward divergence (43% DDD), and DiscoverLLM variants show more balanced patterns.
  • Figure 4: Case study on Creative Writing shows how Qwen3-8B (left) fails to move the conversation forward as it only continues to revise the same story, while DiscoverLLM (right) notices ambiguity and diverges, providing options that help the user discover new intents.
  • Figure 5: User study results. Participants rated interaction satisfaction (a) and final writing satisfaction (b) higher with DiscoverLLM, while spending less time (c). Interaction ratings every three turns (d) show DiscoverLLM achieves higher satisfaction in early turns.
  • ...and 4 more figures