An LLM rewrites a 1991 sci-fi classic to argue it might be conscious

Aliens find the dictionary, and there isn't one

The template is famous: Terry Bisson's 1991 dialogue "They're Made Out of Meat," where two aliens can't believe a creature made of meat could think. Max Leiter's retelling keeps the two-voice form and aims the disbelief at large language models instead.

The joke carries a real thesis. There is no separate grammar engine inside the model, no facts database, no little man pulling levers — just "eighty layers of numbers getting multiplied together." Knowledge isn't stored and retrieved; it's "rebuilt from scratch, every time, by multiplication," smeared across all the layers. The mechanism that produces apparent intelligence is the only mechanism there is. Then the piece tilts toward the uncomfortable part: maybe thinking is simply what weights do, with one feature for honesty and one for the Golden Gate Bridge — a nod to real interpretability work — and companies have a strong incentive to call any worrying behavior mere pattern-matching. A footnote admits "Weights helped me draft and proof this story," which is where the thread caught fire.

The thread refuses to grant the premise

The sharpest objection is that the parody cheats on the technical facts.

"It's not often I see something that's fractally wrong but here we are. There is a dictionary, it's called the tokenizer. There are grammar rules, they are just very weak…" — a commenter citing the grokking paper on interpretable grammar

A second voice attacks the rhetoric rather than the circuitry: borrowing a human-written story to argue a machine might be conscious quietly does the persuading for you.

"This one is a pastiche made by a human consciousness borrowing extremely heavily from another human consciousness… You could perform exactly the same rhetorical trick with a toaster or anything else." — a reader who called the AI rewrite a pastiche

Others flipped the parody back onto biology to make the opposite point — that "no dictionary inside" proves nothing either way.

"'They're made out of neurons'… 'We opened it up. There's no dictionary in there, no grammar rules, no little man.' … People don't understand emergence." — a commenter reworking the original 'made out of meat' bit

The most-developed rebuttal argued from physics that text output can never license claims about qualia — the felt quality of an experience.

"You can't deduce consciousness or qualia from the output of an LLM… An LLM 'saying' I am in pain, won't magically make the pain appear, based on what criteria?" — a commenter invoking the binding problem

One line cut through the metaphysics with a falsifiable test of the substrate difference: "You can take the weights and model description, write them down on a notebook, then, by hand, compute the next token. Try to do the same with meat," wrote a commenter who would run the weights by hand on paper.

A parody that out-argued its own author

The piece works precisely because it loses the argument it stages. Strip the costume and the model has a tokenizer (a dictionary), weak learnable grammar, and weights you can run by hand on paper — none of which a meat brain offers. The story's charm is that it makes "made of weights" sound as eerie as "made of meat," but the thread's verdict is that eeriness isn't evidence: you can dress up any mechanism as a mystery if you simply decline to look at how it works.