I don’t think that using an LLM in a way that acts as a human being should be considered acceptable or appropriate. It should be viewed as detached from reality and concerning due to the mental break from social life that appears to come with such usage.

Given that belief, yes we should be training the models not to mimic emotion.