I agree that chasing score on an Atari game has very limited value, but continuous online multi task learning with sparse rewards and transfer learning is definitely not solved for environments even as simple as Atari. Do you not think those are problems worth grappling with?

→ View original post on X — @id_aa_carmack