It would be the same if the model was "raw", trained only on text completion.
But all models these days are RLHF'ed on (prompt, answer) pairs, so unfortunately they can get confused if the prompt already contains part of an answer.
I think base models are far superior to those boring instruct tuned models. I would rather have a good text completionist than a chat bot. But as far as i know i am in a minority there.
