We can add others, such as BERT (Bidirectional Encoder Representations from Transformers), Mistral 7B, and PaLM (Pathways Language Model).

LLMs apparently train out implicit ontologies in their inner weights, embeddings and the attention matrices. Such ontologies usually take a lot of time to be constructed by humans. A disadvantage of the LLM is here, that the ontologies might not be easily accessed from us to extract it.

If you consider the fact, that attention actually spans up a higher dimensional representations for each term, with each new dimension representing a different meaning of a word, LLMs are in fact digitalisation tools for meaning.

In respect of HMI, they are a more natural approach to interact with computers.