Your question is Handle a Bad Model Surprise. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
You are working on an LLM feature and the model produced an output that was clearly wrong in a way that surprised you. The issue mattered enough that you had to stop and rethink the design before shipping.
Describe a time an AI model surprised you with its output in a bad way. What happened, how did you diagnose it, and what did you change afterward?