Title editorialized for clarity. The issue appears to be reproducible in a sense in the comments. The best theory is that Copilot injects emoji after each bot

Title editorialized for clarity. The issue appears to be reproducible in a sense in the comments. The best theory is that Copilot injects emoji after each bot paragraph outside of the LLM itself and therefore ignoring the user's constraint, meaning the LLM has to attempt to resolve the contradiction.