How would you use an LLM to propose and evaluate an algorithm?
Instruction: Explain your approach, give a practical example, and describe its limits.
Updated
Example Answer
I would describe the problem, constraints, and a baseline, then ask the model for a candidate algorithm with assumptions and pseudocode. I would treat its output as a proposal to investigate. A confident explanation is not a proof of correctness or evidence that the idea is novel.
I would inspect edge cases, compare the proposal with prior work, and verify any complexity claim under explicit assumptions. Then I would benchmark it against the baseline on representative inputs. If the model proposes a faster search method, it must still satisfy the required result quality and handle the difficult cases; a few successful examples would not establish either point.
Concrete example
For a graph search proposal, specify whether edges have nonnegative weights, whether an approximate answer is acceptable, and what changes between queries. Those conditions determine which comparison is meaningful.
Follow-up to practice
What evidence would you need before describing an algorithm as a new contribution?