Comment by Ginden

And, of course, you can train a model to start with repeating prompt in chain of thought!