Resolves to yes if in 1 month I think from my research/the technical paper the openai sora model uses a 3d prior like Gaussian splat or nerf when generating videos.
Related questions
🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ45 | |
| 2 | Ṁ45 | |
| 3 | Ṁ36 | |
| 4 | Ṁ35 | |
| 5 | Ṁ28 |
From this post - "These properties emerge without any explicit inductive biases for 3D, objects, etc.—"
Inductive biases aren't openai's style, they're mostly scale maximalists.
https://openai.com/research/video-generation-models-as-world-simulators
@tftftftftftftftftftftftf The description of 3D consistency as an 'emerging capability' makes me think the answer might actually be NO...
Betting yes on the basis of samples like this: https://openai.com/sora?video=zen-garden-gnome which have weird but perfectly consistent camera motion. I would have expected that if it was just raw pixels, failing to produce reasonable camera motion would also cause distortions in the scene structure.
Not betting very much, because I think it's likely we don't know the answer to this question by the closing date.