I think it is not so much surprising that large language models are particularly good at languages, including programming languages.