btw self-play is the same trick that made AlphaGo better than any human at Go so now that it's finally working on robots with actual bodies, this is probably how we get superhuman robots > self-play means an AI practices by playing against older versions of itself > every time it gets better, its opponent gets tougher too, so it keeps improving without a human ever showing it what to do > but doing it with a body is way harder than Go > because instead of just picking one spot on a board,…

SkildAI@SkildAI4h ago

We trained a robot to play football. How? Self-play for 140 years in a virtual World Cup. Meet the #Messinator: https://t.co/WJNkb2mfoI