Will OpenAI's video model Sora's capabilities be significantly less than the release post represented?
Aug 17

Resolves YES if, in my judgement, the examples release post seems significantly misleading (and notably moreso than the usual hype / selective evidence presentation).

Get Ṁ600 play money
Sort by:

@jacksonpolack How would you judge the impressiveness of the release post? It contained some really good and really bad examples imo.

I found a lot of the examples to be very impressive and much better than any other available video AI? Detail, motion, camera motion, long-term coherence, etc

@jacksonpolack on a deceptive hype-scale, where does the Gemini announcement video rank from 1-10, and what score is needed for this to resolve YES?

(significantly edited)

While that Gemini video was what prompted me to make this, and I do find it misleading, the situation here isn't super analogous IMO - it probably either does generate videos like that when given similar prompts it doesn't.

If those videos were somewhat cherrypicked, I'd still resolve it NO, because it's still capable of doing that and generally if something is 1 in 20 that's very close to 1 in 3 in AI

1 in 500 cherrypicked would definitely resolve YES though.

I'd score that video 5/10 (it mostly wasn't fake! some demos are totally fake!) and say maybe you need 4 out of 10 or higher to get YES but it's a log scale so it's probably not possible to interpret taht

More related questions