Experimental results show that some models, when the pain penalty reaches a certain intensity, proactively abandon the high-score option and instead avoid the penalty.