I discuss something related in the post, and as I said, I agree that if in fact you check the LLM output really hard, in such a manner that you would actually change the text substantively on any of a dozen or a hundred points if the text was wrong, but you don’t change anything because it’s actually correct, then my objection is quantitatively lessened.
I do however think that there’s a bunch of really obvious ways that my argument does go through. People have given some examples in the comment, e.g. the LLM could tell a story that’s plausibly true, and happens to be actually true of some people, and some of those people generate that story with their LLM and post it. But I want to know who would generate that themselves without LLMs. (Also again in real life people would just present LLM’s testimony-lookalike text as though it is their testimony.) The issue with the GLUT is that it’s a huge amount of info, hence immensely improbable to generate randomly. An issue here is that text may have only a few bits of “relevant info”, so it’s not astronomically unlikely to generate a lookalike. Cf. Monty Hall problem; 1⁄3 or 2⁄3 or something of participants find themselves in a game-state where they actually need to know the algorithm that the host follows!
e.g. the LLM could tell a story that’s plausibly true, and happens to be actually true of some people, and some of those people generate that story with their LLM and post it. But I want to know who would generate that themselves without LLMs.
This snippet is my crux here (and perhaps yours?). I do not find it useful to make the distinction between writing a particular sonnet and generating thousands and selecting the same sonnet and only publishing that.
I understand that you’re leaving and think that’s wise and healthy, but am leaving an explanation since I didn’t earlier (expecting a longer conversation).
What I am interested in is the process of a human writing a particular passage somehow. I can imagine a few scenarios and view them as similarly “testimonial”, given that they result in the same text:
A human writes a sonnet “solo”.
A human thumbs through the thesaurus for synonyms for “autumn”.
A human uses a computer to look up “nouns that begin with G”.
A human uses a computer to generate crappy sonnets and splices a few dozen together.
A human uses a computer to generate sonnets and selects one.
I view these as similarly at our current stage of computer sophistication because the intelligence that created the particular sonnet lives inside the human’s head. If they were just prompting Original ChatGPT with “write me a good sonnet”, the result would be bad, but in this article the claim was made that a sonnet we all liked would have its connection to human art and emotion severed if a computer generated it and was edited, and I think that argument proves too much since it seems to have me look down on thesaurus use.
(I will not reply further but really did not want to exit the thread without making my case for my position. I wanted to check whether this was the appropriate case to make or if something else would be more appropriate, but something is better than nothing)
I discuss something related in the post, and as I said, I agree that if in fact you check the LLM output really hard, in such a manner that you would actually change the text substantively on any of a dozen or a hundred points if the text was wrong, but you don’t change anything because it’s actually correct, then my objection is quantitatively lessened.
I do however think that there’s a bunch of really obvious ways that my argument does go through. People have given some examples in the comment, e.g. the LLM could tell a story that’s plausibly true, and happens to be actually true of some people, and some of those people generate that story with their LLM and post it. But I want to know who would generate that themselves without LLMs. (Also again in real life people would just present LLM’s testimony-lookalike text as though it is their testimony.) The issue with the GLUT is that it’s a huge amount of info, hence immensely improbable to generate randomly. An issue here is that text may have only a few bits of “relevant info”, so it’s not astronomically unlikely to generate a lookalike. Cf. Monty Hall problem; 1⁄3 or 2⁄3 or something of participants find themselves in a game-state where they actually need to know the algorithm that the host follows!
This snippet is my crux here (and perhaps yours?). I do not find it useful to make the distinction between writing a particular sonnet and generating thousands and selecting the same sonnet and only publishing that.
That’s just wrong, but I don’t want to prosecute that argument here. I hope you’ll eventually realize that it’s wrong.
I understand that you’re leaving and think that’s wise and healthy, but am leaving an explanation since I didn’t earlier (expecting a longer conversation).
What I am interested in is the process of a human writing a particular passage somehow. I can imagine a few scenarios and view them as similarly “testimonial”, given that they result in the same text:
A human writes a sonnet “solo”.
A human thumbs through the thesaurus for synonyms for “autumn”.
A human uses a computer to look up “nouns that begin with G”.
A human uses a computer to generate crappy sonnets and splices a few dozen together.
A human uses a computer to generate sonnets and selects one.
I view these as similarly at our current stage of computer sophistication because the intelligence that created the particular sonnet lives inside the human’s head. If they were just prompting Original ChatGPT with “write me a good sonnet”, the result would be bad, but in this article the claim was made that a sonnet we all liked would have its connection to human art and emotion severed if a computer generated it and was edited, and I think that argument proves too much since it seems to have me look down on thesaurus use.
(I will not reply further but really did not want to exit the thread without making my case for my position. I wanted to check whether this was the appropriate case to make or if something else would be more appropriate, but something is better than nothing)