Now generate a few candidate responses for the current user to say.

conversation history:
{conversation_history}
user response: