At the earliest stage, when a model has no knowledge, any data can be useful.

“doesn't know anything like when it's zero any data is useful right you can use a lot of video data uh you know in the case of egocentric data i think it's actually a very good way of pre-training robots because human data is so diverse like it covers every single physical activity you can possibly imagine and that gives robot the model such a strong prior on how things how the world can evolve around”