If LLMs will be able to create passable fiction, then I could just cut out the middleman altogether. Instead of paying a human author to create fiction using LLMs, I could just use an LLM to create my own custom fiction.
But, that's a hypothetical problem at the moment. LLMs can't produce compelling fiction because they don't have a well-developed theory of mind.
If LLMs will be able to create passable fiction, then I could just cut out the middleman altogether. Instead of paying a human author to create fiction using LLMs, I could just use an LLM to create my own custom fiction.
But, that's a hypothetical problem at the moment. LLMs can't produce compelling fiction because they don't have a well-developed theory of mind.
That's my point. But, similar to the Turing test, creating fiction doesn't require sentience, consciousness or the practical application of psychology or philosophy. We already happily anthropomorphise animals and robots. It just requires a machine to win a place at the 'imitation game' table. And 'compelling' will continue to remain in the eye of the beholder-cum-buyer.
". . . creating fiction doesn't require sentience, consciousness or the practical application of psychology or philosophy."
Well, technically you're right. LLMs *can* create fiction of some kind. The question is how many people will enjoy reading it. An entertaining novel involves suspense, psychological insight, and unexpected twists and turns. The author has to anticipate the reactions of the readers to some extent, which requires a sophisticated theory of mind. An A.I. program of some kind might be able to do this someday, but it probably won't be an LLM.
I've tried using ChatGPT-4 to write fiction and the results weren't that interesting. ChatGPT is very good at mimicking prose style (e.g., it can produce stuff that sounds like D.H. Lawrence or Edgar Allen Poe), but the writing had no real direction; it just cranked out descriptive paragraphs that didn't tell a story or lead anywhere.
I'm sure somebody will find a way to monetize LLM fiction and it might even be lucrative, but I doubt that LLM-produced novels will rise to the top of the best-seller list any time soon.
But their potential use in automating the publishing pipeline to replace human copy editors, proofreaders etc is no doubt being explored by companies looking to cut costs.
If LLMs will be able to create passable fiction, then I could just cut out the middleman altogether. Instead of paying a human author to create fiction using LLMs, I could just use an LLM to create my own custom fiction.
But, that's a hypothetical problem at the moment. LLMs can't produce compelling fiction because they don't have a well-developed theory of mind.
That's my point. But, similar to the Turing test, creating fiction doesn't require sentience, consciousness or the practical application of psychology or philosophy. We already happily anthropomorphise animals and robots. It just requires a machine to win a place at the 'imitation game' table. And 'compelling' will continue to remain in the eye of the beholder-cum-buyer.
". . . creating fiction doesn't require sentience, consciousness or the practical application of psychology or philosophy."
Well, technically you're right. LLMs *can* create fiction of some kind. The question is how many people will enjoy reading it. An entertaining novel involves suspense, psychological insight, and unexpected twists and turns. The author has to anticipate the reactions of the readers to some extent, which requires a sophisticated theory of mind. An A.I. program of some kind might be able to do this someday, but it probably won't be an LLM.
I've tried using ChatGPT-4 to write fiction and the results weren't that interesting. ChatGPT is very good at mimicking prose style (e.g., it can produce stuff that sounds like D.H. Lawrence or Edgar Allen Poe), but the writing had no real direction; it just cranked out descriptive paragraphs that didn't tell a story or lead anywhere.
I'm sure somebody will find a way to monetize LLM fiction and it might even be lucrative, but I doubt that LLM-produced novels will rise to the top of the best-seller list any time soon.
True, they're more of a text generation accelerator/author's assistant at the moment, as with this account of Sudowrite usage: https://www.thecreativepenn.com/2023/02/08/how-i-used-generative-ai-tools-for-my-short-story-with-a-demons-eye/
But their potential use in automating the publishing pipeline to replace human copy editors, proofreaders etc is no doubt being explored by companies looking to cut costs.