Here's a human-made video of a Rube Goldberg machine:
https://www.youtube.com/watch?v=Ss-P4qLLUyk
Eliezer Yudkowsky has predicted that the ability to generate a video like that (indistinguishable from reality even on close human inspection) from a prompt will require an AI that's within 1 year of AGI.
Since this is another leading indicator, I thought it would be good to have Manifold market for it.
Yudkowsky's thinking is that it takes a high-fidelity physics model to make all the balls and dominoes and things interact realistically. If you understand the physical world at that level, AGI isn't far behind.
FAQ
1. How long does the video have to be?
Maybe a minute? Don't quote me on that yet. Let's discuss in the comments what best hews to the spirit of the question, about a leading indicator for AGI.
Ask other clarifying questions in the comments as well -- I'll add them to this FAQ!
Related Markets
https://manifold.markets/dreev/will-ai-be-lifechanging-for-muggles
https://manifold.markets/dreev/instant-deepfakes-of-anyone-within
https://manifold.markets/dreev/in-what-year-will-we-have-agi
[Ignore any auto-generated clarifications below this line. I'll add clarifications to the FAQ above.]
1,000
3.00This would be a nice market if there was a proper way to resolve. As it is now this will just attract traders on vibes and probably some smart trades based on knowing creator's previous actions.
I suggest a different format, like "If [some test for the validity of the claim] is performed by the end of 2028, will it show [some criterion an AI generated video needs to achieve]?". Like resolving based on a poll where one must guess which of 2 videos is AI generated and responses must be within a tight margin around 50:50. Which of course opens up new problems, as impressive Rube Goldbergs are widely known...
Maybe "Will an AI generated video of a Rube Goldberg machine reach 1 million views on Youtube without any commenter expressimg doubts about human origin?" Maybe the Youtube Partner Program can be operationalized in some way... But I doubt such easy resolution criteria are possible, as Youtube will only use their no-monetization-for-AI-content for boring stuff and spam.
Anyways, I think it's very hard to come up with proper criteria for Yud's prediction which would he suitable for a market.
@Primer Thanks for the suggestions. I think having different variants of markets is really valuable so I encourage you to create one or more yourself.
Currently we're not at all in a gray area on this one -- it's a clear NO if we were resolving today. It's entirely possible it will be an obvious YES in 3 years. So I think betting now based on vibes is not irrational. For the somewhat unlikely case that we do end up in a gray area at the end of 2028, traders should ask clarifying questions.
PS: I've started an FAQ in the market description. The first one is about how long the video needs to be. Next questions could include things like how close of scrutiny the AI-generated video has to hold up for. Like can you slow it down, replay it, try to replicate it in the real world?