FREE IDEA! Take a human-authored codebase from <2022, have teacher LLM create a "sloppy" reimplementation of a single feature, then train student LLM to undo the slop. Should bias away from broad exceptions, blank returns, attribute/type checking, or low-q defensive coding.
Training Student LLMs to Fix Sloppy Code Reimplementations
By
–
