Will this Yudkowsky tweet on AI video generation hold up in 2024?
➕
Plus
761
Ṁ960k
Dec 31
34%
chance

Same as this market but closes and resolves end of 2024
https://manifold.markets/journcy/will-this-yudkowsky-tweet-hold-up



"This market resolves YES if at close (end of **2024**) my subjective perception is that this was a good take (https://twitter.com/ESYudkowsky/status/1563282607315382273) --e.g., AI-generated video really is that good--and NO if it seems like Eliezer was importantly wrong about something, e.g., AI-generated video still sucks, or still couldn't be the cause for serious doubt about whether some random moth footage was made with a camera or not."

Get
Ṁ1,000
and
S3.00
Sort by:
bought Ṁ500 YES

@Duncn the conjoined twins convinced you?

It's easy to forget how bad models were in 2022. I would bet more, but I'm afraid of this resolving on a technicality like the models can't count to 15 or generate 60s+ in one shot, rather than the actual quality of the video output which was the interesting part of this question when it was created and is absolutely there if you're not a lepidopterist.

bought Ṁ5 NO

cool 5 mana no on this because if you this tweet holds up u got a real wacky view on the world fr

@RiskComplex yeah this confirms my feeling lol

@NoahRich I bet :)

@RiskComplex me when i can't decipher the flow of time from a video: OMG THIS LOOKS AMAZING!!!!!!

@NoahRich Like a moth flapping its wings?

It resolves YES because even if you are able to guess it is most likely not AI, it is not so far outside of Sora as to be impossible. Therefore, your first thought should be to think carefully if it is or is not generated. One could imagine the moth video being the result of multiple Sora shots where any weird and poor results have been discarded. Even if it can’t consistently produce this, it may be close if enough attempts are dumped into it and the best footage is edited together.

I have no bets in this market.

bought Ṁ250 YES

@LiamZ I've changed from NO to YES, the technical capability certainly is not there yet, but I think a majority of Manifold users will do what the tweet and the spirit of the market implies and that is at least think if any given video couldve been made by AI.

@ScipioFabius personally this market was too subjective for me and I’ve been burned on very subjective markets before. IMO this should have at least resolved with a poll so traders could estimate the “you” population profile. I wish you luck.

@LiamZ I don't agree because I've simply never seen a video from any of these Gen AI video tools that is anywhere nearly as coherent as the linked moth video, and so I don't see people openly skeptical of videos like it.

@DanW if you show random people this https://x.com/minchoi/status/1866191030829084698

And then ask them if they think the moths could be made by the same process, what percentage do you think would say yes?

I would guess over 50% but this whole market is subjective like I said.

@LiamZ In that case the percentage would probably be very high. But once you also show them cases of Sora failing to generate simple videos of four legged animals walking, the percentage would probably drop quite a bit.

As soon as there is any significant amount of motion involved it becomes immediately obvious that the video is AI generated.

The monkey clip is the only one from that thread which manages to escape the uncanny valley.

@4rthurRainbow it’s subjective and I wish you luck. As mentioned, I have no bets.

I think stopping to wonder and evaluate the probability the video is generated by thinking through how close the motion looks to canny vs uncanny confirms the world posited by the tweet. Sora can produce canny motion of some animals but not consistently.

@LiamZ Since when has a typical human's first thought ever been "to think carefully" about anything? You think I thought carefully about making this comment? No! I am a stochastic parrot. A parrot!!

There is a conceptual space between YES and NO here. It is possible that both Eliezer's take was NOT importantly wrong about something, but that by the end of 2024 video models just aren't that good yet.

Also his tweet was in August 2022 and he said "2-4" years. So it seems it would be more fair to wait till August 2026

You're right, but this market is specifically about it having held up in 2024, so I think that's how it should resolve.

@CornCasting Fundamentally, Eliezer's question is whether or not you'd question if a video like that moth video was generated by AI after seeing it. https://x.com/minchoi/status/1866191030829084698

@RiskComplex @dflz The problem still is that @GeorgeVii could at the end of 2024 say "We don't have anything that can produce moth pics that good but it looks like it could happen in 2026 so I say the tweet holds up"

This market resolves "NO if it seems like Eliezer was importantly wrong about something, e.g., AI-generated video still sucks, or still couldn't be the cause for serious doubt about whether some random moth footage was made with a camera or not."

With the new SORA model released today I think this is case-closed. (These are just the first examples from the first 1-2 hours after its released):
https://x.com/minchoi/status/1866191030829084698
https://x.com/minchoi/status/1866191027482100011
https://x.com/edwinarbus/status/1866192355763900798
https://x.com/minchoi/status/1866191028958466144

@RiskComplex first one is short enough you can’t see flaws. The rest, no so much.

@JaimeSantaCruz I completely agree about the first one, but what is also important I think is that it doesn't even do anything new. It definetely has a video just like this in its training data, and it has just copied it.

It's a big improvement over previous models but longer and more complex videos do consistently become incoherent.

@JaimeSantaCruz The tweet doesn't say it will be impossible to tell the difference, but that your "first thought" will be oh I need to actively discriminate.

@JaimeSantaCruz I agree. The question is whether or not if you saw a video of that moth flapping its wings would you ask yourself whether or not its AI generated. I think it's clear that at this point the answer is that of course you would.

If you look at that video and think "This is certainly not AI" you either don't realize what the SOTA is or you've voted NO lol.

@DanW the moth video in question is a series of short shots of different moths. It's relatively simple and doesn't need to be generated in a single request.

@LiamZ

if the AI's prompt was "beautiful video of 15 different moth species flapping their wings, professional photography, 8k, trending on Twitter"

The tweet is explicitly about the combination of moth shots from a single request. That has been a large part of the NO bettors' reasoning, or at least mine—though I think we're in the clear for 2024 even if were a single moth shot.

@Jacy If you read the tweet very literally, it was about one specific prompt. But I don't think that's the spirit in which the market is intended. If there existed an AI right now which could generate that exact moth video, but only if you added "pretty please with a cherry on top" to the end of the prompt, would you still argue that this should still resolve NO?

The actual market is on whether George perceives the tweet as a "good take". In my reading, the tweet is mostly about no longer being able to trust that videos are real. If new technology is good enough and cheap enough that popular, real-looking videos on social media today are often fake, that seems like enough to resolve YES.

@Jacy You can use the same prompt repeatedly. I have no bets but that's not good reasoning about the tweet itself regardless of if you think models are here or not.

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