Neural Nets will generate coherent 20-min films by the end of 2025 that win film festival awards [See link to better qu]
➕
Plus
104
Ṁ27k
2025
16%
chance

Set criteria:

  • Must be coherent

  • Must have coherent scenes and plot/plots/themes

  • Must win some film festival awards which we'd expect humans to win

  • Cannot be heavily edited by humans. I guess a few cuts or some lighting is allowed but heavy editing and this is perhaps doable already

  • Must not just be winning cos of novelty

Inspired by tweet thread:

Link: https://twitter.com/RichardMCNgo/status/1640568776495353860?s=20

Get
Ṁ1,000
and
S3.00
Sort by:
predicts YES

I'm sort of minded to resolve this N/A to free up your mana. I think @firstuserhere's market is just better.

Currently I won't because norms, but if it were just me and copies of me I think i would.

@NathanpmYoung If people want to free their mana, they can sell. I'm not invested anymore, with N/A I would lose my profit. I really care about these 7 mana.

predicts NO

@Zardoru @NathanpmYoung +1 to the notion that an N/A resolution would unfairly remove well-earned profits. (Disclaimer: I also have a little of those.)

Two markets for shorter duration films, for shorter timelines.

predicts YES

I'm duplicating this market and removing the criteria for winning awards, to see how the market feels about just the capabilities to do this

predicts YES
predicts YES

@firstuserhere Good call

predicts YES

@NathanpmYoung Could you pleases provide clarity about the resolution criteria that you intend to have as well as the one you intended at the time of the market creation? The points raised about coherence not really meaning it'll be award meaning are extremely valid. "Must not just be winning cos of novelty" combined with other conditions makes it incredibly likely to resolve to NO, despite the title's conditions "coherent 20-min films" being met

Must win some film festival awards which we'd expect humans to win

This is a really tough condition. I feel like a lot of people are betting just based on the title, which is a lot weaker.

predicts NO

@MatthewBarnett Agreed, that condition is so much more difficult for a NN than just generating somewhat reasonable looking movies

predicts NO

@MatthewBarnett it's an insane condition. A lot of outputs of mj are very very good, and they could easily have won contests ten years ago. Now? no, no one will award such a project when it's trivial to make 1000 full quality analogs in 10 minutes.

@MatthewBarnett Absolutely ridiculous condition given the title. MOST coherent films made by humans aren’t award winning.

Oh wow, this is a really surprisijg condition. The title should almost certainly reflect the strength of it in some meaningful way

predicts NO

And also note that a single award is insufficient!

predicts YES

Easy you lot, @firstuserhere was right to just duplicate this. It took me about a whole morning to write these up as it was. I'm gonna make mistakes. If you want better markets, write 'em yourselves.

@NathanpmYoung Thank you for your work on this and other markets! Taylor Swift has it basically right, though she couldn't find a good rhyme with "At least you'll be rolling in Mana"

predicts YES

@MatthewRitter I think this is true external to one's community but internally I'd say "I will push back against haters because I'm not the only one that pays the costs"

@NathanpmYoung Just clarifying if needed - my criticism of any specific conditions in any market doesn’t take any appreciation away from the efforts you put into your markets ❤️

Not betting on this because it's underdefined what actually counts as "a neural net generating a film". Must it be a system that outputs a .mp4 file when given a text prompt describing the film? So not "it wrote a script which was then turned into a film by students" or anything like that. Also, must it be capable of generating multiple distinct films, in any style and on any subject it's asked to?

It's very hard to properly formulate bets about such open-ended AI capabilities.

Are human brains neural nets?

For the record, my object-level prediction of this is ~30%, though much hangs on the criteria "Must not just be winning cos of novelty." I expect AI will produce really interesting minutes-long videos (especially if humans are allowed to do some editing), but I don't expect it to win anything like a Turing test against human-produced 20-minute films.

predicts NO

Whether these "draft criteria" are going to be used would change the probability by a lot.

"Standard of degree level film student" sounds to me like "not totally boring". Can someone confirm?

Ngo:Gpt-4::Gary Marcus:GPT-2

© Manifold Markets, Inc.Terms + Mana-only TermsPrivacyRules