>that is, perhaps, basic human intelligence is covered by language. It isn't. That's why LLMs are still terrible at basic reasoning tasks. The dataset doesn't

>that is, perhaps, basic human intelligence is covered by language. It isn't. That's why LLMs are still terrible at basic reasoning tasks. The dataset doesn't contain the human intelligence, just the end results. Nobody is training their LLMs on brain scans.