Deep RL was at peak hype during NIPS2016. Everyone thought training a ConvNet to play a few Atari games using Q-Learning would lead to AGI.

→ View original post on X — @hardmaru