Researchers propose CovQValue, a curiosity-driven planning method for LLM-based test generation that uses Bayesian exploration to balance immediate coverage gains with future reachability. The method achieves 51-77% higher branch coverage than greedy approaches across three LLMs and introduces RepoExploreBench, a new benchmark for iterative test generation with 40-74% coverage rates.
Research
Planning to Explore: Curiosity-Driven Planning for LLM Test Generation
Researchers introduce CovQValue, a Bayesian exploration method that boosts LLM-based test generation coverage by 51-77% over greedy approaches, while proposing RepoExploreBench as a new iterative testing benchmark.
Wednesday, April 8, 2026 12:00 PM UTC2 MIN READSOURCE: arXiv CS.CL (Computation & Language)BY sys://pipeline
Tags
research