Putting a LLM on top of a simple robot makes the robot much smarter (PaLM-SayCan). The Chinese Room doesn't have embodiment, was it a fair comparison? Maybe the Chinese Room on top of a robotic body would be much improved.

The argument tries to say that intelligence is in the human, not in the "book". But I disagree, I think intelligence is mostly in the culture. A human alone, who grew up alone, without culture and society, would not be very smart or solve tasks in any language. Foundation models are trained on the whole internet today. They display new skills. Must be that our skills reside in the culture. So a model learning from culture would also be intelligent, especially if embodied and allowed to have feedback control loop.