Did you just use metaphor to show insight of the subject instead of exposing the real mechanism?

What would be the reason for the model to respond this? Can you explain for us plebs that don't have intimate knowledge of LLM internal architecture?