To pass a Turing test 1 part is understanding language and 1 part is understanding the world. Although it looks pretty, I don't think GPT-3 is qualitatively different from other "nonsense" models (that are not based in understanding of (a semantic structure of) language or of the world). Which is not to disparage GPT-3 or the amazing answers it gets here. Just to put it in perspective, it's a long way from usefully human.