Reinforcement learning does not scale. Efficient environmental sampling does. One aspect of intelligence is the ability to predict the world while observing so little. #deeplearning #ai
To observe so little implies that there's an imaginary simulation that is able to extract new information without interacting with the environment. In short, a game is played between imagination and reality.
Deep Learning has revealed to us what methods work very well in reality. This leads to imagined new architectures that will be exponentially more capable than the previous architectures. #deeplearning #ai #agi
