Where is this going?

The new model announcement states that the new model intentionally has a 2% chance to deviate from the user's preferred response style. This is a step away from what makes ChatGPT useful, to me. Not sure where this is taking us, but I hope there will be a way to disable this "feature."