Game playing is not LLM. They’re game-specific reinforcement learning models. It’s not easy, but definitely doable with existing tech. Sony’s GT Sophy is a good demonstration on what they’re capable of.
Machine learning is not viable for anything other than simpler 2d games or small segments of more complex games. The training required to get good results on that is intense already.
It is expensive, but it does work. We’ve already seen things work to a limited extent on StarCraft 2, Dota, and Gran Turismo, and those are all multiplayer games. The article seems to be talking about single player games, which simplified things a lot.
Game playing is not LLM. They’re game-specific reinforcement learning models. It’s not easy, but definitely doable with existing tech. Sony’s GT Sophy is a good demonstration on what they’re capable of.
Machine learning is not viable for anything other than simpler 2d games or small segments of more complex games. The training required to get good results on that is intense already.
It is expensive, but it does work. We’ve already seen things work to a limited extent on StarCraft 2, Dota, and Gran Turismo, and those are all multiplayer games. The article seems to be talking about single player games, which simplified things a lot.