Yup, this is what I’m talking about. I’ve been following googles WANNs and other methods for reducing the size of these “strategies” drastically. In the WANN pa

Yup, this is what I’m talking about. I’ve been following googles WANNs and other methods for reducing the size of these “strategies” drastically. In the WANN paper they have an example of going from About 2500 Parmams to 40. I’ve got my own hunch as to where it is going, but I’m wondering if openai is studying this at all. Like there are simple primitives in civil engineering like the level and pulley, I expect we find a zoo of tiny primitives which are the things that these agents are learning.