Ask Before You Optimize: Dynamic Pre-Formulation Clarification for Interactive Optimization
The paper tests whether optimization agents know when a problem statement is too incomplete to model.
It introduces OR-Clarify, a benchmark where agents must recover hidden formulation details through limited interaction with a simulated user. The benchmark measures whether agents ask useful questions, stop at the right time, avoid silent assumptions, and manage interaction cost. The authors also propose InterOPT, a two-stage framework for identifying formulation-critical gaps before deciding whether to ask again or proceed. In choice-based experiments, InterOPT beats all baselines on exact slot recovery; in open-ended tests, it stays competitive with strong prior methods. ArXiv · AI/CL/LG's note
It introduces OR-Clarify, a benchmark where agents must recover hidden formulation details through limited interaction with a simulated user. The benchmark measures whether agents ask useful questions, stop at the right time, avoid silent assumptions, and manage interaction cost. The authors also propose InterOPT, a two-stage framework for identifying formulation-critical gaps before deciding whether to ask again or proceed. In choice-based experiments, InterOPT beats all baselines on exact slot recovery; in open-ended tests, it stays competitive with strong prior methods. ArXiv · AI/CL/LG's note
score 5