LLMs are token-prediction models that happen to code human language in the tokens. You could train a similar model based on e.g. sensor inputs for moderating a plant and regulating a grid.