So that being said, does this mean the new model of ChatGPT 3.5 is less steerable than before using the system prompt?