UTGen teaches large language models to generate effective unit tests that both identify bugs and predict correct outputs without needing gold solutions.

• Addresses the key trade-off between error detection and output prediction
• Optimizes test generation for maximum debugging effectiveness
• Enhances LLM code-debugging capabilities through better test creation
• Improves software quality assurance through automated testing

For engineering teams, this research offers a pathway to more efficient debugging workflows and higher code quality with less manual testing effort.

Learning to Generate Unit Tests for Automated Debugging