@welchlabs: Do our AI models really need to be generative?

Richaz :

The result is a model that can’t be used for generative tasks efficiently 🙄 For example for robotics, you have to get a goal frame somewhere, so you’re very constrained. Furthermore, to generate a single sequence of actions you have to run the predictor hundreds of times on randomly sampled actions. I’m not sure for we can deal with these problems. Any ideas?