Depends on the model. Some just go into “immediate mode” and just do whatever you ask, others operate fine but have trouble with tasks/tools. While others will go down a quant that was basically neglected since inception and you get garbage back. Random chars or endless loops.