Neural Nets will generate coherent 20-min films by the end of 2025 (less strict criteria market, $1000M subsidy)

Set criteria:

  • Must be coherent

  • Must have coherent scenes and plot/plots/themes

Inspired by tweet thread:


Get Ṁ600 play money
Sort by:
bought Ṁ0 of YES

DALLE-3 was likely trained with some sort of iterative feedback mechanism with the intention of making it follow prompts more tightly, and it worked. That indicates that the hardest plausible bottleneck to a YES resolution is less difficult than I previously expected.

Video generation is already sophisticated (and this market doesn’t require photorealistic or Turing-test-passing video), and coherent scripts are way easier to produce than good ones. Several companies seem to be expressing investment interest in projects like this, so the funding’s there to put everything together.

A poll with different percentage ranges for this market. If you'd like to make a prediction but don't wanna trade on it, use this free poll to lock it in. It closes at the end of 2024, 1 year prior to this markets resolution

sold Ṁ20 of NO

It is worrying to me that on such a subjective market the market creator holds one of the highest YES shares. Especially since there is nothing in the resolution criteria about the quality of the film.

bought Ṁ0 of NO

@FlorisvanDoorn reputation damage prob not worth the Ṁ407

then again, i would bet all these markets down to <10% if i was sure of no shenanigans

predicts YES

@FlorisvanDoorn Sigh, I wish Manifold had a better way for someone to participate in their own markets, while securing the interests of all the traders. After all, I am very interested in this market and it's evolution over time.

That said, I'll sell my shares from this market, and not trade again. I'll also subsidise the market more so as to safeguard a bit more against sudden movements.

sold Ṁ197 of YES

@nickten can you share which markets you're referring to?

@FlorisvanDoorn I sold my shares, and won't trade on this. I'm creating a poll where I'll mark my prediction. That seems like a decent alternative. Perhaps an attached poll should be a toggle for some markets, and creators can toggle it on if they don't wanna trade.

bought Ṁ0 of NO

@firstuserhere search for "AI". that's it. that's the list.

jk, but it does feel like a lot people on this website are very optimistic about major AI advancements soon™ and might be eager to bend definitions so it "counts". i haven't actually seen any serious markets be "misresolved" so my fears are prob silly.

also leagues disincentivize betting long term markets to much away from 50%

@nickten haha nice one. At this point I don't care much about profit maximizing, so I'm not at all hesitant to resolve against my position. I've done my fair share of profit maximizing and don't get much value in that anymore. So, assume the risk for my markets to be minimal, or consider my positions as emotional hedges against a lot of gains in capabilities.

@firstuserhere I understand the impulse to trade on your own markets, since you often make markets about the things you care about. But it shows a good character that you decide to sell your shares.

Just to be clear: I wasn't accusing you of planning to rig this market. I was just considering that this market is so ambiguous that the following two wild positions are almost(!) compatible with the current market description:

  • Current neural nets are already sufficient to resolve this YES, since there are ways to generate a 20 minute very crappy film where so little happens that it isn't positively incoherent.

  • Neural nets would have to be superhuman at generating films to resolve this YES, since most films have (minor) inconsistencies.

I expect that @firstuserhere would choose a pretty reasonable position within this grey area, but it is harder to do this if you have a stake yourself (you might be looking at candidates with YES-colored glasses).

predicts YES

Created two markets for shorter timelines asking for shorter duration of coherent films

Thanks for duplicating. I think yours is better

bought Ṁ20 of NO

Does it have to be an end to end neural network? Or there can be a separate neural network that writes script?

predicts YES

@NikhilVyas Can be separate, no problem. Given the AI-generated script, the next AI system creates coherent 20-min film that tightly follows that script. Must be able to generate 20 min long films though.

bought Ṁ0 of YES

This is a duplicate of the following market with a bit less strict criteria

@firstuserhere this one doesn’t have the condition for a film to be accepted at a film festival

@long yes that's right. The linked market does, this one doesn't.

More related questions