Tests on typical inputs miss wrong LLM repairs; add one edge case
A code repair that runs cleanly can still return wrong results, and in a preregistered study the inputs of the...
Read More