Humans are "multi-modal". Sure we get plenty of non-textual information, but LLMs were trained on basically every human-written world ever. They definitely see many orders of magnitude more language than any human has ever seen. And yet humans get fluent based after 3+ years.