
Interactive Debugging for LLMs
Enabling AI to explore and debug code like humans
debug-gym creates a text-based environment allowing LLMs to interactively explore codebases, overcoming the limitation of fixed context windows.
- Provides a Python debugger interface for LLMs to inspect variables and trace execution
- Enables models to gather relevant information by navigating through code
- Supports development of agents that debug interactively rather than relying solely on static context
This research matters because it bridges the gap between LLMs' coding capabilities and real-world programming scenarios, where iterative debugging and exploration are essential engineering practices.
debug-gym: A Text-Based Environment for Interactive Debugging