Episode 29: Jim Fan, NVIDIA, on foundation models for embodied agents, scaling data, and why prompt engineering will become irrelevant
Generally Intelligent
00:00
Is There a Way to Scale Up the Training Process?
One of the main weaknesses of my clip is the horizon of the task. We can only do like up to eight to 16 seconds of video and then do this contrast of learning. So there is no way we can build a house using this model because of the context lines. Of course, you cannot fit the training data in the context then it will not be able to do long horizon. But back then we really hope like building a house is kind of our major milestone to do but that turns out to be much harder than we expected.
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
