Optimization by PROmpting (OPRO) has shown that large language models (LLMs) can generate candidate solutions for combinatorial problems through natural-language prompts. However, the abstract for a new arXiv paper notes that OPRO's optimization process is difficult to track, which prevents LLMs from being fully trusted in iterative decision-making.

The paper, titled "Think in Strategies, Act with Tools: Traceable Agentic Optimization for Combinatorial Problems," proposes a framework that makes the process more transparent. By integrating explicit strategies and external tools, the method aims to give each optimization step a clear rationale and a verifiable action, rather than relying on opaque prompt-driven outputs.

This shift toward traceability could make LLM-based optimization more practical for real-world combinatorial tasks, where understanding why a solution was chosen is as important as the solution itself. The abstract does not provide experimental results, so the effectiveness of the approach remains to be seen, but the direction addresses a known limitation of OPRO.