Google, OpenAI, and Anthropic are competing to see whose AI can play Pokémon the best — Twitch streams of beloved RPG game test the models’ true might

Twitch streams of different AI models playing old Pokémon games have garnered hundreds of thousands of comments as the audience watches robots try to emulate human thinking. Unlike Pong or its derivatives, a game like Pokémon is much more complex, requiring risk assessment and logical reasoning, all great metrics to test and improve an AI.