AI models are not proficient in debugging and cannot replace human coders in this aspect, according to researchers at Microsoft Research.
Microsoft Research has developed a tool called debug-gym to evaluate and enhance AI models' debugging capabilities. It provides AI models with access to debugging tools that were previously unavailable for them.
Researchers found that, without the use of debug-gym, AI models perform poorly in debugging tasks. Even with the tool, they are still far from matching the expertise of experienced human developers.
Although AI has been integrated into varios aspects of software development, it is not yet ready to replace human coders for debugging purposes.