LLM's spit out derivative nonsense. Highly convincing nonsense, but derivative nonetheless. Any true originality in the output still comes from the human engineering the prompt.
Basically, LLM's know how to follow the rules, color within the lines. A real artist understands those rules well enough to break them and create something novel.
Attempts to rigorously study the relative creativity of LLMs have already found them more creative than average humans: https://www.nature.com/articles/s41598-023-40858-3
They're only getting better, fast.
If you're not convinced, what GPT4-level (text or multimodal) or Midjourney-level (imagery) models have you tried, and how much?
I think filmmakers like Daniel Kwan and Daniel Scheinert (Swiss Army Man, Everything Everywhere All at Once) are where Hollywood should be focusing their attention, and their sweep at the Oscars this year has me cautiously hopeful that at least a few people in power are paying attention.
This is what I see as the problem. Hollywood needs to stop relying so heavily on derivative goop and give actual creatives more leeway in creating something original.
You can only ride old IPs for so long before people lose interest and stop buying tickets, as Hollywood has been learning recently. You need people coming up with fresh novel ideas if you ever want to come up with the next big thing.
Some executive asking Chat-GPT to spit out a new Marvel movie script isn't going to do that.
This idea that AI cannot be creative but humans can is something that begs for a test; and that test needs to be applied against humans, as well. If AI can pass it at least as often as humans can, then AI is at least as creative as humans; within the bounds of that test.
As it happens, an AI has recently passed such a test[0].
0: https://www.technologyreview.com/2023/09/14/1079465/ai-just-...
I don't see why you couldn't have an AI do the same, by training it towards a particular aberration in output until that became something it understood well.
But at this very moment, AI generating anything suitable for something other than SEO blogspam isn't possible anyway. It might be right around the corner, or it might be "right around the corner" like self-driving cars were in 2015.