Only NNs of infinite size or precision. Under more realistic assumptions, transformers without chain of thought are actually limited in what they can solve: https://arxiv.org/abs/2207.00729