As a programmer that works indirectly on DNNs and sometimes LLMs. No. I can’t actually do that. (disclaimer: my work is low-level, I work with a the level of the stream processors with memory and vector acceleration for DNN, most of that work applies to more general purpose computing. so it’s not a pure AI job)
I can setup a system where you can feed it data to make it do the thing you want, sometimes. But that’s on you as the trainer, not on me, the programmer. I don’t really have a way to write an LLM to express artistic intent. As you can train it to do wildly different things that I never imagined someone might try.
A friend of mine who writes a lot of fiction has been playing with LLMs and found some things they are reasonably good at, especially when you need some feedback.
Throwing ideas at them and asking them to riff on what you’ve provided can help trigger another burst of creativity, early on - but they are only good at broad brushstrokes. There’s still a lot of legwork needed by the author.
Asking it to examine blocks of text looking for cliches and well-worn tropes was surprisingly effective; and another useful experiment was to get the LLM to adopt the persona of an editor and critique a piece for style and structure.
It no way replaced the sort of feedback you’d get from a knowledgeable person, but it was helpful in those limited cases. Especially in the wee small hours.
