Will this be used for literally anything aside from making slop "games" in year or two? People in the video are talking about this being so called "world model" where it can simulate realistic scenarios and training AI agents (in this case presumably robots), but how true is this?

The bald fucker at the head of DeepMind kept talking about how VEO 3 is physics engine because it generated realistic-looking water physics, but that means jack shit. Image generators in 2022 were capable of doing realistic-looking lightning while having completely 0 spatial awareness, which you could see with prompts like "Red cube on a bigger blue cube". And in VEO 3 you can see it in any action scene where things just happen with no rhyme or reason, all of the physics stuff is just hallucinated. So if all of these environments are mostly either just the user walking through some generic environments or the model hallucinating actions, are they even remotely useful for any sort of robot training?

Only use case is porn and of course you can't use it for that

>>106158832 (OP)
In the future you will have full dive like vr with worlds that can be generated real time based on your neural inpulse.

They will make the world so bad that people will live in pods and escape in fantasy.

>>106159068
Don't worry anon, people still need to have a job and stuff so they cant spend 24/7 in AI generated fantasy. And UBI wont come, elites will just cull everyone who isn't a billionaire with major AI company holdings.

>>106159068
I hope not because otherwise I'd be mad that I have wasted like a third to half of my life already

>>106158832 (OP)
Google has been quiet about how much expenditure it takes to make these simulations run for 1 minute. And they said nothing about the cost of these runs. However they mentioned "autoregressive" in their documents, which means quadratic complexity as time goes on. That means the probability of these runs collapsing as time goes on approaches 1.
There's only so much you can simulate in the span of a minute.

>>106159368
>expenditure
by this I meant compute.
>the cost of these runs
By this I meant $

Ai is advancing so fast, it's unbelievable. "vibe coding GTA clone" doesn't seem unreachable anymore. Gaming industry is not ready.

This does nothing of value for humanity.
ai was a mistake.

(No offence to sentient machines, i'm referring to the slop people turn out with LLMs).

>>106158832 (OP)
>where it can simulate realistic scenarios and training AI agents (in this case presumably robots), but how true is this?
It seems possible, easier to train in simulate physical environments with various combinations & permutations of different variables. Rather than having to physically create such an environment for training.

>>106159835
>Implying the cyberpunk vr gooning sessions this will eventually lead to is of no value to humanity

Demis is right about the physics engine part. Turns out if you just throw enough data at the models they'll reverse engineer physics:
https://arxiv.org/abs/2506.17774

>>106159958
So when can we expect robots that arent clumsy morons?

>>106158832 (OP)
>One step closer

>>106158832 (OP)
So is this anything else but an AI generated HDRI map through witch you move?