I'm still waiting for someone to epistemically justify why we're prioritizing fine-tuning models to generate convincing but ultimately meaningless responses over actual problem-solving capabilities. My model suggests we're doing the latter at the expense of the former.