I suspect you could train a model to just shut up and follow instructions. I.e. instead of "Do X -> Sure! As a large language model, I'd love to help you with X!", just "Do X -> X".
This would avoid giving the model a dull default voice. But it wouldn't be sufficiently hedged and "safe".
It's like instruct = Gain cosistentcy, lose creativity. Base = No consistency (really), but creative.
Could a model be trained or positioned to be in the middle of these two?