What if LLMs are just basic synthetic models of human brains trained only on lots of language?

The output depends on the input and training. Input stuff and reinforcement train what you do/don't want more of.