What does a random bug in one LLM's frontend app have to do with learning how to do prompt engineering well?