What makes this actual "reasoning" by the LLM, versus a document-extender that generates text where human readers perceive fictional characters as doing reasoning?<p>We can evoke the same kinds of document-characters which discuss loving their children, but that doesn't mean the children are real or that the computer has learned to love.<p>I harp on this distinction a lot, because I feel it's a common and fundamental mistake in these discussions and in product marketing.