Google just released a video generator that is a ball hair away from perfection. The hallucination rate from their latest models is <1% and dropping you just see cherry picked screenshots.
Only sometimes, with enough generations you can already make indistinguishable videos for the most part. You’re seeing these mistakes because it’s amateurs spending $100 not professionals spending $10k.
Given current rate of pace Veo4 should be out in a couple months.
I was referring to the demos shown at I/O, my brain seems to see them all as “flat”, it’s like a reverse magic eye thing. It reminds me of the uncomfortableness of a fever dream. I may be the exception tbf!
Thanks! These are all great, the details, everything is very clear and crisp compared to before, hardly any garbledy goop, but I can’t shake the vertigo idk
I don’t think image generators are really in the same category though. They’ll have their applications but they’re not going to be a fundamental change to society the way AGI will be.
Yes it does. It’s one component of a broader system. The ability to generate helps it interpret. An AGI might use a diffusion model to imagine scenarios, generate visual plans, or process sensory input.
The AGI, by definition, will make something vastly better than diffusion model. That’s one of the cornerstones of AGI, it will explode it’s capabilities.
The statement assumes AGI must surpass all narrow models in every domain, but this is not a requirement by definition. AGI is defined by its generality across tasks, not by superiority in each specialized field.
Google just released a video generator that is a ball hair away from perfection. The hallucination rate from their latest models is <1% and dropping you just see cherry picked screenshots.
The physics and perspective are still horrible, every video makes me want to vomit my brain
Only sometimes, with enough generations you can already make indistinguishable videos for the most part. You’re seeing these mistakes because it’s amateurs spending $100 not professionals spending $10k.
Given current rate of pace Veo4 should be out in a couple months.
I was referring to the demos shown at I/O, my brain seems to see them all as “flat”, it’s like a reverse magic eye thing. It reminds me of the uncomfortableness of a fever dream. I may be the exception tbf!
I showed this one to my friend and she said ‘But they faces aren’t AI generated, right?’
https://www.reddit.com/r/aivideo/comments/1kxs6yq/were_cooked_a_zerocost_ai_demo/
There’s a bit at the end where the spaghetti disappears, the chef walks away a bit quick while still speaking, but otherwise it’s nearly flawless.
For scenes with lots of action and complex physics it’s still very noticeable
https://old.reddit.com/r/aivideo/comments/1ktgjwh/pushing_veo_3_to_the_limit/
But it’s already good enough to replace several scenes in blockbusters. Dream scenes, cut-aways, etc.
Look at this sausage dog
https://xcancel.com/nmatares/status/1924931844879134804
It even gets the audio right when it moves between hardwood and carpet.
Thanks! These are all great, the details, everything is very clear and crisp compared to before, hardly any garbledy goop, but I can’t shake the vertigo idk
I don’t think image generators are really in the same category though. They’ll have their applications but they’re not going to be a fundamental change to society the way AGI will be.
It’s part of AGI and will be a massive shift. They are to video what punk was to music.
Agi and image diffusion has literally nothing in common though?
Yes it does. It’s one component of a broader system. The ability to generate helps it interpret. An AGI might use a diffusion model to imagine scenarios, generate visual plans, or process sensory input.
The AGI, by definition, will make something vastly better than diffusion model. That’s one of the cornerstones of AGI, it will explode it’s capabilities.
The statement assumes AGI must surpass all narrow models in every domain, but this is not a requirement by definition. AGI is defined by its generality across tasks, not by superiority in each specialized field.