MULTI-AGENT COOPERATION

Summary from the TLDR AI newsletter

Building on Google's Transformer architecture, the authors proposed training sequence-model agents against many different opponents so they learned to adapt within each game without hardcoded assumptions about how others learn.