Research
StoryCoder: Narrative Reformulation Yields 18.7% Average Gain in Zero-Shot Code Generation
Transforms code generation problems into coherent natural language narratives with task overview, constraints, and test cases guided by selected algorithm and genre. Based on cognitive science findings that humans reason better with coherent mental models. Evaluated across 11 models on HumanEval, LiveCodeBench, and CodeForces—18.7% average improvement in zero-shot pass@10. From NAVER AI Lab.
Source
↳ Follow the thread