00:00 Speaker A
Now, one of the most important things we have to do when it comes to physical AI is to first create the data to train the AI. Where does that data come from?
00:09 Speaker A
How do we teach an AI the ground truth of physics, rather than having it have a language because we've created a bunch of text that we think of as ground truth for the AI to learn from?
00:20 Speaker A
There are plenty of videos, but not nearly enough to capture the variety and types of interactions needed.
00:30 Speaker A
And this is where brilliant minds come together to transform what used to be computing into data.
00:43 Speaker A
Currently, using synthetic data generation that is conditioned on the laws of physics and grounded and conditioned by ground truth,
00:54 Speaker A
You can now selectively and intelligently generate data that can be used to train your AI. For example, what goes into this AI is this Cosmos AI world model on the left.
01:06 Speaker A
Traffic simulator output.
01:10 Speaker A
Now, this traffic simulator isn't nearly enough for the AI to learn.
01:17 Speaker A
This can be incorporated into the Cosmos foundation model to generate physically based and physically plausible surround video that AI can learn from.
