Movie = Any interactive experience that can be considered a continuation or successor to 20th and early 21st century movies
High-quality AI = at least as capable as ChatGPT 4 or DALL·E 3
Prompt = Any command, verbal, written or neural
Suppose there are interactive movie-like experiences, where the viewer can influence what happens (like Netflix's Bandersnatch but good).
The market resolves YES if, in the course of such experience, the consumer can prompt a high quality AI into being and ask or have characters ask it things. The consumer must be able to specify the nature of the AI in a variety of ways. For example, any of the following prompts should work:
a chatbot similar to ChatGPT 3.5
image generating AI that only draws animals
ChatGPT 4 but super skilled in mathematics
It must be the consumer of the movie who has the AI created and not the producers, It does not matter how the movie goes about implementing this, it is fine if it just pretends to be the AI requested as long as it does it well.
See also:
1,000@ProjectVictory Cool project
The key point with regards to your question is
The consumer must be able to specify the nature of the AI in a variety of ways
Jailbreaking to access the underlying model is a valid method, but the viewer must also be able to specify what kind of AI should be brought into being. It should be possible to bring into being most variations of ChatGPT 4 or DALL·E 3. It must not break the continuity of the game/movie experience.
If a game/movie uses a model that can satisfactorily emulate variations of ChatGPT 4 or DALL·E 3, and you can get it to follow, e.g. "From now on this character behaves like ChatGPT 4 who only talks about animals" (and other variations) and then are able to query it, that is enough to resolve this market.
@admissions do you basically accept any game as being an interactive movie for your question's purposes?
@Ramble No, that's too far. I took "Game prototype Inworld Origins is arguably an interactive movie." at face value, but originally I wrote
> Any interactive experience that can be considered a continuation or successor to 20th and early 21st century movies
I will require multiple external sources comparing the game to a movie or describing it as movie-like. I think many games fit this criterion, but definitely not all or most.
This is so stupid, but also legitimately seems very likely
If we got a system composed of ChatGPT writing a script + Sora making a video based on the script + some audio generator making audio based on the script and video + some UI that asks the user "what should happen next," we'd basically already be there
@CDBiddulph Damn, the fact you said we'd basically already be there a year ago is crazy, it always surprises me how much many AI-pilled people overestimate current progress. The reason to be scaling-pilled is exponentials, sufficient quantity = quality all of its own, discontinuous effects, eventual RSI slow takeoff, etc.
@DavidHiggs I still think this could exist in principle without any major technical breakthroughs.
The main technical challenge is adding audio to a video, but the fact that this hasn't really happened probably has more to do with the labs not prioritizing it than it being inherently difficult. You don't even need the audio if you allow the dialogue of your "movie" to happen via subtitles.
Probably the main reason something like an AI choose-your-own-adventure hasn't gotten big is because Sora is still too slow for it to be very engaging.
Once you have this, there's no reason that the AI couldn't write a script for the "fictional AI" that imitates another language model or image model, as long as that model is somewhat less capable than itself.
@CDBiddulph my point was that Sora or other SotA is not at the level of generating coherent, continuous, object permanence obeying, multi-character interacting videos of anything at all, that is comparable to movie footage. Even for 15s.
Let alone doing all the other stuff, stitching it together coherently, and interacting in real time, to a single prompt. We’re not currently close, let alone 12 months ago
@DavidHiggs ChatGPT can totally write a coherent script though, which can be split into chunks for Sora to render, maybe setting the last 5 seconds of the previous clip as the context for the next clip. And if all you need is a movie scene with an AI talking through a computer terminal in a single room, that could be very easy to render.
@DavidHiggs Your market doesn't mention any requirements about coherence, anyway. If in the middle of this movie scene with the AI, the camera pans to reveal an object that wasn't there 15 seconds ago, that shouldn't affect the outcome of this market as long as the behavior of the AI itself matches the user's prompt
@CDBiddulph it’s not my market, but the resolution description seems to strongly imply quality requirements: continuation or successor of 20th/21st century movies implies at least as good/better. Also, it gives a reference point: Netflix’s Bandersnatch but good.
And of course this market is a reference to several other markets with various requirements about high movie quality, so if there’s any doubt it seems obvious which way to lean as far as interpretation.