Comment by BoorishBears

It's no mystery: solving LLM voice requires making the model worse at other things they value above all other things