> Who cares about the implementation? I guess one difference between LLM-style AI and most other systems is that when other systems produce the wrong output yo

> Who cares about the implementation? I guess one difference between LLM-style AI and most other systems is that when other systems produce the wrong output you can usually step through the code, figure out why, and be confident that the same issue won't happen again.