Here’s an interesting story about a failure being introduced by LLM-written code. Specifically, the LLM was doing some code refactoring, and when it moved a chunk of code from one file to another it changed a “break” to a “continue.” That turned an error logging statement into an infinite loop, which crashed the system.
This is an integrity failure. Specifically, it’s a failure of processing integrity. And while we can think of particular patches that alleviate this exact failure, the larger problem is much harder to solve.
Davi Ottenheimer comments.