It's going to be difficult to judge quality. Current ones are interesting for just a few videos (no more than 15 minutes total), and then you don't want to see any more ever. For the first 3 minutes they are high quality, then it's all the same slop, subjectively - that's my experience at least.

I predict that the same problem will exist by the time this market resolves, however I don't know whether the resolution criteria would resolve to "yes" in that scenario.

My prediction: maybe they will be good for up to a few hours of total watching (or less), then you'd have to wait for 2029 to enjoy them again.

Surely this is going to resolve yes? Or do fine tunes not work if the training distribution is heavily censored?

Not sure if I would call it high quality, but it's apparently high quality enough right now to make the frontpage of Pornhub...

@Siebe Creator is banned, let's hope resolution will be straightforward. Not even sure if this market is allowed according to Manifold's terms and conditions.

Q1: how long a movie would count as 'full'?

Q2: I expect ai-generated movies to be ~structurally different from what movies exist now, so: what counts as a movie? (Should it have verbalizable plot, dialogue, voice acting, consistent-ish characters, ~direction? Is Samsara (2011) a movie in the spirit of the question?)

Q3: I expect most if not all ai art to be considerably human-directed, so: how much/which parts of the movie must be ai-generated to count? (I assume >80% of the video is a must and that normal-for-today amount of curation and touchups of the raw content is fine; unsure whether human-supplied plot, dialogue, voice acting, and scene-edition are in spirit.)

@797a

1. Let us take a threshold of 8 minutes. This is the duration that the average person spends on p**nhub.

2. That is correct. I did not see the recommendation yet, but judging from the remainder of the sentence, I'd agree.

3. Yes that assumption holds fine to my taste (touchups by humanity are fine, it is the core material which is generated by the AI)

"Normal-for-today" editing is still a lot of work. ChatGPT 03-mini-high estimates 50-100 employee hours are needed to convert raw footage into an 8 minute, high-quality pornographic movie.

Are we going to resolve the market YES if a human has to put in 50-100 hours behind the scenes?

I am not familiar with the censorship on manifold and do not know how strict the moderation is on any links or explicit words. Hence I will link to a hackernews post that shows one example of work in this area: https://news.ycombinator.com/item?id=32572770

Yes @Joshua these do need to be photorealistic

Do you generally endorse all of Scott's criteria?

"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."

@Joshua Yes I do, but it did not seem appropriate (primarily due to my unfamiliarity with rules on Manifold re:NSFW content) to write a variant of that in the description (ex: what's the equivalent of star wars?!!) Otherwise, I do not know what ... "big budget studio" would mean, because some (ex: me) prefer home-made content - so this part seemed more subjective to me, and I did not include it.

@sixtynine ... but yes, high-quality would mean high-quality. In AI video generation, it seems harder to generate a 480p video than a 1080p video from my current understanding.

Cool, thanks for clarifying!

At first glance this seems mispriced because one would assume this is a subset of "in 2028 will AI be able to generate a full high quality movie to a prompt", but since the "quality" of NSFW movies is such a lower bar, it actually makes sense.