dalelane’s Twitter Archive—№ 19,854

Fascinating paper on using generated worlds to train reinforcement learning systems worldmodels.github.io/ It's a long read, but so worth it - very well explained, helped by plenty of interactive animations and video demos