Meta = intelligence
The more you abstract the more you understand
(less intelligent) Structure -> Language -> Meaning -> Pattern (more intelligent)
Even largest language models still sometimes struggle with human level pattern understanding. I don't think its about size anymore. Its probably about lack of compute or wrong type of training (data).
Patterns and meaning are the meta of language. Its something you cannot clearly see from language, so its between the sentences. The invisible hidden in the visible.