If you had a robot doing all the computations of a LLM on pencil and paper instead of GPUs, would anyone think it was conscious?