There are no proofs that language models wouldn't be able to become generally intelligent or achieve AGI, and Bach doesn't see obvious limitations they would run into, though he finds them aesthetically inelegant and computationally inefficient.