The field of language models is witnessing a significant shift towards understanding the nuances of human language interaction. Recent studies have investigated the lexical diversity of language models, revealing that they do not produce human-like texts in this regard. Furthermore, research has shown that language models can influence human language use, with some studies indicating a convergence between human word choices and language model-associated patterns. The direction of the field is moving towards a deeper understanding of the mechanisms underlying language model behavior and their impact on human language. Noteworthy papers include:

• A study on lexical diversity in language models, which found that newer models produce less human-like texts than older models.
• Research on language change, which discovered a significant increase in the usage of language model-associated words in unscripted spoken language.
• An investigation into linguistic convergence, which revealed that language models can adapt to the linguistic patterns of their users, but often overfit relative to human baselines.