chess, shogi, and Go https://en.wikipedia.org/wiki/AlphaZero https://arxiv.org/html/2409.12272v2
poker https://www.science.org/doi/10.1126/science.aay2400
StarCraft 2 https://en.wikipedia.org/wiki/AlphaStar_(software)
Dota 2 https://en.wikipedia.org/wiki/OpenAI_Five
factorio
https://fxtwitter.com/_neel_kant/status/1920516491025482066
pokemon https://x.com/sundarpichai/status/1918455766542930004
minecraft https://x.com/BasedBeffJezos/status/1831017904613228598
snake https://www.youtube.com/watch?v=PJl4iabBEz0
https://www.youtube.com/watch?v=zFLQU70QstY
https://www.youtube.com/watch?v=NUl6QikjR04
ai x trackmania
reinforcement learning beats real time humans, finding some sota strategies on its own, but needing reward pushing for other ones, and it doesnt beat the creativity of tool assisted speedruns by humans, which might be just for now? it also had input accuracy limitations
i wonder where could DeepMind get to if they threw 7 trillion dollars onto this problem