Why would it require a director if a generative process can use the information it has on audience (even individual) preferences to produce the story and format that will best hook consumers?
Because it will perform similarly to the way that writing LLMs do…. They obey and produce singularly predictable and droll stories that have trouble keeping the attention of a five year old. It’s the definition of stale tropes and predictable scenes at its worst.
It has no idea of the mind of the viewer or the reader. It’s literally generating the most probable next few words to tack on to the story.
LLMs are great for a lot of stuff, but they are by design not at all creative in any admirable sense of the term. They cannot produce a narrative that is simultaneously unique or genuinely surprising without it also being nonsensical.
Hallucinations are not a bug, they are a result of proper operation , just undesirable. The same thing that makes nonsensical“hallucinations “ if the “temperature” is set too high is what prevents llms from having any unique ideas if the temperature is set low enough to not hallucinate wildly.
LLMs are text prediction engines. Extremely useful and revolutionary in many ways, but not in creative work. Everything they do is by definition derivative and likely.
What graphics models do in images is no different, it’s not all that creative, and works much better when you -tell it- to be derivative. It’s just that the nature of graphic representation is mainly predictable and derivable… so it doesn’t bore us when it produces derivative, predictable work.
It has no idea of the mind of the viewer or the reader. It’s literally generating the most probable next few words to tack on to the story.
LLMs are great for a lot of stuff, but they are by design not at all creative in any admirable sense of the term. They cannot produce a narrative that is simultaneously unique or genuinely surprising without it also being nonsensical.
Hallucinations are not a bug, they are a result of proper operation , just undesirable. The same thing that makes nonsensical“hallucinations “ if the “temperature” is set too high is what prevents llms from having any unique ideas if the temperature is set low enough to not hallucinate wildly.
LLMs are text prediction engines. Extremely useful and revolutionary in many ways, but not in creative work. Everything they do is by definition derivative and likely.
What graphics models do in images is no different, it’s not all that creative, and works much better when you -tell it- to be derivative. It’s just that the nature of graphic representation is mainly predictable and derivable… so it doesn’t bore us when it produces derivative, predictable work.