Nvidia trained a model called NitroGen on 40,000 gameplay videos from 1,000 games to teach robots control policies, achieving 52% success. The question: can massive video datasets replace traditional robot-specific training data, the way internet text scaled up language models?
Basically there's Nvidia's introduced new model known as nitrogen which is even Those trained on 40,000 videos from Twig and YouTube across I guess around 1,000 games from which the success rate was about 52% which is quite cool. And this they say it's similar to the robot training as well. So the main question here is that Nvidia's nitrogen seems to learn control policies from massive gameplay videos. So is this robot pin without robot data?
So one thing what happened with large gen models were that we had this incredible vast amount of data repository in the form of Internet and we just. That's just the way these language models work in a way that just made sense and gave us the illusion of intelligence. Similarly, I mean every company is trying to find out is can something serve as something serv as Internet for robots? Just like the Internet served to language models, is there, is there terms of videos in terms of some content that just we can feed to these robots and they behave just like humans, right? So for example, in terms of cognitive labor, you've seen that large.
FDE Epi : 16 | AI Agents, Billion-Dollar Infrastructure & the Future of Intelligence
