Reproduce “Ask Before You Optimize: Dynamic Pre-Formulation Clarification for Interactive Optimization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sihan Ge, Yichen Lin, Chenyu Zhou, Jianghao Lin, Tao Yao, Dongdong Ge
Ge et al. from Shanghai Jiao Tong University and Cardinal Operations introduce pre-formulation clarification as a distinct task for LLM-driven optimization, focusing on identifying and recovering "formulation-critical facts" before model generation. They developed the OR-Clarify benchmark and the InterOPT framework, which substantially improves the exact recovery of problem requirements in choice-based interaction settings and reduces silent assumptions.
Key insights
- Pre-formulation clarification is a challenging problem for LLMs; off-the-shelf models struggle to achieve high exact recovery and frequently make silent assumptions.
- Explicitly identifying, tracking, and persistently managing a ledger of "formulation-critical facts" through dynamic gap search is essential for comprehensive problem understanding.
- Structured, choice-based interaction protocols, when combined with gap-guided questioning, significantly enhance the exact recovery of problem requirements and reduce silent assumptions, indicating the value of guided interaction for information elicitation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.