Are humans just LLMs in meat suits? Arturo Nereu doesn't quite think so, but in a recent essay, he lays out the uncomfortable parallels between how large language models work and how he experiences his own thinking. The more he interacts with AI tools for research and development, the more he notices the overlap — and the more he wonders whether the comparison goes deeper than metaphor.
Nereu has read over 300 books in the past five years and remembers almost none of them in detail. Once he picked up a book at a store, started reading, and only realized halfway through that he'd already read it. The knowledge is somewhere in there, shaping his decisions, but he can't trace which idea came from which source. He also hallucinates: his memory of his first day at work has been retold so many times that the details have drifted from whatever actually happened. Even his vision is a kind of hallucination — only the words next to his cursor are clear; the rest is filled in by his brain.
And he can be prompted. Be nice to him, build a bond, and he'll behave differently than he would otherwise — the human version of jailbreaking an LLM. Nereu notes that humans have always mapped their brains onto whatever they've just invented: clocks, steam engines, hydraulics, and now AI. He's not convinced he's an LLM. But he's not entirely convinced he isn't.
Previously:
