Large Language Models were invented and trained by engineers, with little input from any previous study of language. But of course there was 100 years of study of the complexities and mystery of natural language before LLMs were even thought of, and some of this previous work may be valuable.
I will give a brief and informal talk on pragmatics, which is the theory of how we can mean more than we actually say. A consideration of pragmatics also reveals the unconscious thought that we use to understand the intended meaning of language.
I cannot (yet) argue that an understanding of pragmatics is directly useful in designing or using large neural language models, but if you are interested in modern natural language processing, it seems good to have some acquaintance with existing theories of language and meaning.