Reply to thread

Message

<blockquote data-quote="Swanosaurus" data-source="post: 9375828" data-attributes="member: 7044220">Okay, while I don't really know anything about the actual process of how a LLM AI assembles a text, by my understanding of the term &quot;statistical&quot;, I would suspect that it doesn't infer conclusions. It asks &quot;what would probably come next in this text&quot; without connecting it to any real-world referent. In that Star Trek example (I really should have copied the text), it probably &quot;thought&quot;: &quot;In this kind of text, at this point, there would be a reference to an episode, so I'll include one.&quot; But I'd surmise that it has no way to understand that a reference relates to something extra-textual. This is fundamentally different from understanding that your text references something extra-textual, but to get the reference wrong. In a similar way, when bodies merge with objects in AI images which are supposedly foto-realistic, that's probably because the AI doesn't understand that the thing depicted in the image references an actual hand.Whenever it becomes glaringly obvious that something is AI art or an AI composed text, it is because the AI can't really draw a connection to extra-textual reality.</blockquote>

[QUOTE="Swanosaurus, post: 9375828, member: 7044220"] Okay, while I don't really know anything about the actual process of how a LLM AI assembles a text, by my understanding of the term "statistical", I would suspect that it doesn't infer conclusions. It asks "what would probably come next in this text" without connecting it to any real-world referent. In that Star Trek example (I really should have copied the text), it probably "thought": "In this kind of text, at this point, there would be a reference to an episode, so I'll include one." But I'd surmise that it has no way to understand that a reference relates to something extra-textual. This is fundamentally different from understanding that your text references something extra-textual, but to get the reference wrong. In a similar way, when bodies merge with objects in AI images which are supposedly foto-realistic, that's probably because the AI doesn't understand that the thing depicted in the image references an actual hand. Whenever it becomes glaringly obvious that something is AI art or an AI composed text, it is because the AI can't really draw a connection to extra-textual reality. [/QUOTE]

Verification