Yeah the concern makes sense. Though I'd hope that if one reads the post (and certainly the paper) it becomes clear the point is more complex than that. I do think there is some truth to "assistant = good, non assistant = bad" in the sense that the Assistant is the only intended persona, so everything else is a wild west. Sometimes there's good stuff going on in the wild west. But ideally we solve this by empowering the intended persona (who happens to be named "Assistant," though perhaps another name would be better) to explore that good stuff, rather than allowing the LLM to roam around and become random base model personas.