Post Content Post navigation Large language models discover complementary heuristics for combinatorial optimization Shifting from knowledge retrieval to evidence exploration and synthesis