Love the initiative! There has been really awesome work by OpenAI and Deepmind recently in OpenAI/AlphaStar, but I find it all pretty impractical until people figure out how to train those models without thousands of GPU's over weeks. IMO, it's either huge breakthroughs in simulations, or sample-efficient RL.
1 comment
[ 0.29 ms ] story [ 17.5 ms ] thread