LLMs Can Design Near-Optimal OR Algorithms
A single untuned LLM query produced OR algorithms competitive with specialized methods in the tested problems.
Jackie Baek tests inventory control, queueing network control, and assortment optimization. The paper separates solving one instance from writing an algorithm before seeing the evaluation instances. Its strongest tested model, gpt-5.6-sol, matched or beat the best existing method on almost all evaluated cases. The author frames frontier LLMs as a serious empirical baseline for well-specified operations research algorithm design. ArXiv · AI/CL/LG's note
Jackie Baek tests inventory control, queueing network control, and assortment optimization. The paper separates solving one instance from writing an algorithm before seeing the evaluation instances. Its strongest tested model, gpt-5.6-sol, matched or beat the best existing method on almost all evaluated cases. The author frames frontier LLMs as a serious empirical baseline for well-specified operations research algorithm design. ArXiv · AI/CL/LG's note
score 6