Episode 29: Jim Fan, NVIDIA, on foundation models for embodied agents, scaling data, and why prompt engineering will become irrelevant
Generally Intelligent
00:00
Using a Learning Reward Function for Image Generation
The mind club itself is learned. Like the other reward functions like collecting diamond, there are obvious like shaping reward to it because to collect diamond, you will need to craft like wooden tools and stone tools. So we can still do that for many of the tasks, but for some of the creative tasks, which means that you don't have kind of a ground troops success criteria for these tasks, that you will have to use a learn reward function. And examples are let's say building houses, right? Building a console, like what makes a castle a console? I think this is a very similar problem as image generation. Let's say we want to generate a picture of a cute cat,
Transcript
Play full episode
