EG "make me a 120 minute Star Trek / Star Wars crossover". It should be more or less comparable to a big-budget studio film, although it doesn't have to pass a full Turing Test as long as it's pretty good. The AI doesn't have to be available to the public, as long as it's confirmed to exist.
1,000People are also trading
List of predictions for autonomous Tesla vehicles by Elon Musk - Wikipedia
Why would his predictions about AI movies be any more accurate? The correlation with inaccuracy is obviously not chance.
@DavidHiggs maybe a feature length period drama about a man going through all the stages of grief until he finally accepts that he has to eat a shoe
@ScottAlexander maybe someone has asked this already, but the prompt can be as long as wanted? Like a "Make me movie that follows this script: [whole script here]" for example. Or is the question also about a generating a sensible script?
@egroj The question provides an example of a prompt that must lead to a high quality movie being generated.
@Yakushi12345 I was the one who made that comment but the question is not "will there exist at least one prompt that outputs a good movie". The question is "will it output a good movie for all / most prompts"
@pietrokc Hum. "Be able to <...>" seems like it might be operationalized as "show an example for which <...>". I don't think a fair reading of this question is "In the space of all possible prompts for movies, will there exist an AI model that can turn all of them into good movies? And will this AI exist by 2028?". Granted, I also don't think a fair reading is "Will an AI be able to turn one specific prompt into a good movie, and fail to do so for all other prompts?", but there's a wide space between those extremes.
@inaimathi I agree this market should resolve based on "reasonable" prompts like the one given in the extra text after the title question. If you prompt the AI "make me a bad cowboy movie" it wouldn't be fair to count that as resolving the market NO.
However I think a lot of flexibility should still be required on the acceptable prompts. Otherwise I can trivially train an "AI" that just memorizes ~1000 existing good movies, and for a given prompt just returns the one closest to the prompt.
"One of the things that I am confident will exist in about two years is a Sora 2 style model that can run on a MacBook without copyright, personal opt-in, or other safety filters."
One interesting thing to think about is that nobody has yet come close to generating a "pretty good" novella either (ie, one that anyone actually wants to read). It requires a similar level of narrative sophistication and consistency, but it's way cheaper and you don't have to worry about visual artifacts, etc.
I would expect it to happen at least ~6 months before the movie.
@n_t
>One interesting thing to think about is that nobody has yet come close to generating a "pretty good" novella either (ie, one that anyone actually wants to read).
given the number of ai-generated novels on amazon, someone must be reading them. My personal experience is that the best AI models are somewhat above the level of writing the worst fanfic writers. Certainly it is possible for an AI to write a better screenplay than say Avatar 2 or SW Ep IX.
@n_t It should be obvious that you have no way of knowing this. Similar to AI generated images where people saw crappy ones and thought it was the SOTA, not realizing that some of those 'real' photos they were comparing to were actually AI generated.
@jim Note that in order for this to resolve YES, the movie effectively has to be made in 2027, otherwise Jan 2028 it will resolve NO.
Elon always predicts things long before they happen, usually many years before. So if he says AI will make a good movie in 2027, that is very good evidence this market will resolve NO.