Code Debugging and Refactoring with OpenBugger: a Python Library for Self-Supervised AI Debuggers
summary
OpenBugger is an open-source Python library that uses LibCST to perform invertible AST substitutions, turning correct code into buggy variants and back again so labels are exact and synthetic. The pipeline parses 2,000 LeetCode solutions, injects structural bugs (including some NumPy/Pandas patterns), and yields roughly 20,000 pairs plus automatic debugging traces. Data generation is static (no execution), supports chaining bugs, and includes checks that the regenerated syntax tree matches the intended fix; a GitHub branch is available though LibCST 1.0 may affect pip installs.
AI-written summary · about
chapters (10)
- 0:00Introduction and motivation
- 5:42What are language models
- 10:01How GPT models work
- 13:31Recent AI model breakthroughs
- 15:35OpenBugger’s debugging data approach
- 16:38Concrete syntax trees and libcst
- 18:47Generating debugging traces
- 21:58Why synthetic data matters
- 23:00HK3 lab projects overview
- 25:09Q&A debugging and inversion
related talks
50mIdentifying and fixing Issues in Code using AI based tools
35mHello Bug, My Old Friend
51mProgramming is Debugging
06mHow Automated Tools Can Communicate Effective Strategies for Fixing Bugs
29mPowered by AI: A Cambrian Explosion for C++ Software Development Tools
29mLiberating the Debugging Experience with the GDB Python API
40mFrom Bugs to Brilliance: Leveraging AI in Modern Software Testing
26mDetecting and Mapping Code Defects Introduced by Generative AI Using CodeLogic & Neo4j