I agree with you that it’s a good idea to carefully select the promise to test the prompt template. Makes sense to create a list of some interesting promises to try.
I think the best prompt template will work for most premises, however ridiculous. Moreover, the ridiculous ones are unusually useful for testing the prompt, as they demonstrate the true strength of the GPT’s imagination: most human writers will fail to write a good story about a woman whose head suddenly became a tomato fruit.
BTW, below is the first chapter generated with the following promise:
GuySrinivasan, a user of Lesswrong website, is shocked by an AI-generated story about a woman whose head suddenly became a tomato fruit. He advises the author to try to get a prompt which writes a great story given a reasonable premise instead.
GuySrinivasan sat in the dim light of his study, fingers dancing across the keyboard, the glow from his screen reflecting off the dark circles under his eyes. The world outside his window was silent, but inside the room, the quiet hum of the computer tower was a soothing companion to his thoughts. The smell of strong black coffee mingled with the musty scent of old books lining the walls, creating an atmosphere that was both homely and academic.
He was a regular on Lesswrong, a website dedicated to the discussion and development of artificial intelligence, rationality, and the future of humanity.
This evening, like countless others, he was absorbed in the latest discussions. A debate about the ethics of AI was gaining traction, but a particular thread caught his eye. It was a story, AI-generated, about a woman whose head turned into a tomato. The narrative was as ridiculous as it was engrossing.
He let out a short chuckle, the sound echoing in the silence. “Well, this is new,” he muttered, running a hand through his greying hair.
For all the absurdity of the plot, he found himself drawn into the story. The narrative was laced with subtle humour, its characters painted in hues of relatable human emotion, and the world it was set in had its own unique charm, even if that charm was rather bizarre.
Guy found himself musing on the power of AI. It could generate stories that evoked laughter, even from absurd premises. Yet, he wondered, could it create a masterpiece from a more… reasonable premise?
Moreover, the ridiculous ones are unusually useful for testing the prompt, as they demonstrate the true strength of the GPT’s imagination.
Pretty sure writers care way more about craft than “imagination”. Use a reasonable premise to figure out what causes the LLM to spit out good writing, then throw in the unreasonable premise to see what happens, if you like. Anyway that’s theorycrafting. I’ll post a candidate prompt and one-shot results for my reasonable premise, the tomato premise, and the meta-absurdity premise.
I agree with you that it’s a good idea to carefully select the promise to test the prompt template. Makes sense to create a list of some interesting promises to try.
I think the best prompt template will work for most premises, however ridiculous. Moreover, the ridiculous ones are unusually useful for testing the prompt, as they demonstrate the true strength of the GPT’s imagination: most human writers will fail to write a good story about a woman whose head suddenly became a tomato fruit.
BTW, below is the first chapter generated with the following promise:
GuySrinivasan sat in the dim light of his study, fingers dancing across the keyboard, the glow from his screen reflecting off the dark circles under his eyes. The world outside his window was silent, but inside the room, the quiet hum of the computer tower was a soothing companion to his thoughts. The smell of strong black coffee mingled with the musty scent of old books lining the walls, creating an atmosphere that was both homely and academic.
He was a regular on Lesswrong, a website dedicated to the discussion and development of artificial intelligence, rationality, and the future of humanity.
This evening, like countless others, he was absorbed in the latest discussions. A debate about the ethics of AI was gaining traction, but a particular thread caught his eye. It was a story, AI-generated, about a woman whose head turned into a tomato. The narrative was as ridiculous as it was engrossing.
He let out a short chuckle, the sound echoing in the silence. “Well, this is new,” he muttered, running a hand through his greying hair.
For all the absurdity of the plot, he found himself drawn into the story. The narrative was laced with subtle humour, its characters painted in hues of relatable human emotion, and the world it was set in had its own unique charm, even if that charm was rather bizarre.
Guy found himself musing on the power of AI. It could generate stories that evoked laughter, even from absurd premises. Yet, he wondered, could it create a masterpiece from a more… reasonable premise?
Pretty sure writers care way more about craft than “imagination”. Use a reasonable premise to figure out what causes the LLM to spit out good writing, then throw in the unreasonable premise to see what happens, if you like. Anyway that’s theorycrafting. I’ll post a candidate prompt and one-shot results for my reasonable premise, the tomato premise, and the meta-absurdity premise.
Edit: https://www.lesswrong.com/posts/goq3HfxZaQcomuH6d/usd300-for-the-best-sci-fi-prompt?commentId=nG99eewTtxrPM2sEk