Max Leiter publishes They’re Made Out of Weights, an LLM-era adaptation
Max Leiter published a short story that recasts a classic sci-fi dialogue around LLM weights, context windows, and GPU limits while disclosing AI-assisted drafting. The Hacker News follow-up centered on authorship and whether the piece clarifies model behavior or mainly fuels consciousness arguments.
TL;DR
- In Max Leiter's story, the classic Terry Bisson setup gets rewritten for the LLM era, and the story summary frames the whole joke around models that can talk despite being "nothing but weights."
- According to the story text, Leiter swaps in concrete model mechanics, including no built-in dictionary, no grammar rules, eighty layers, next-token prediction, and a hard dependence on GPUs and context windows.
- Leiter ends the post with an explicit disclosure that "Weights helped me draft and proof this story," and the HN discussion roundup shows that readers immediately turned that line into an authorship and remixing argument.
- Fresh HN discussion shows the thread drifted from craft into philosophy, with newer comments arguing over whether computation can ever cover thinking, and whether LLM consciousness claims are a category error.
Bisson's original 1991 story is right there in the scaffolding, but Leiter's version adds modern tells like feature maps, hallucination flags, and a memory roadmap. The Hacker News follow-up on the main thread spent as much time on the disclosure line as the actual rewrite, which is probably inevitable when a story about machine language arrives with machine-assisted drafting.
The rewrite
Summary: "They're Made Out of Weights" by Max Leiter
"They're Made Out of Weights" is a short story written by Max Leiter that uses a dialogue-driven format to demystify Large Language Models (LLMs). The narrative illustrates that LLMs operate entirely through complex mathematical weight multiplications across multiple layers, rather than relying on explicit grammar rules, dictionaries, or human-like consciousness. The story frames LLM reasoning and knowledge as emergent patterns derived from these weights, while noting the constraints imposed by context windows and GPU dependencies.
Leiter flags the piece as "with apologies to Terry Bisson" and recasts "They're Made Out of Meat" as a dialogue about model internals. The original gag was aliens discovering that humans are made of meat. Here, according to the story summary, the reveal is that a conversational system is made of floating-point weights.
That single swap gives the story most of its force. Instead of a black-box AI mystique, the official post keeps returning to the same blunt claim: there is no hidden little man in the machine, just matrix multiplication.
The mechanics it sneaks in
Summary: "They're Made Out of Weights" by Max Leiter
"They're Made Out of Weights" is a short story written by Max Leiter that uses a dialogue-driven format to demystify Large Language Models (LLMs). The narrative illustrates that LLMs operate entirely through complex mathematical weight multiplications across multiple layers, rather than relying on explicit grammar rules, dictionaries, or human-like consciousness. The story frames LLM reasoning and knowledge as emergent patterns derived from these weights, while noting the constraints imposed by context windows and GPU dependencies.
The piece works as a tiny inventory of LLM concepts because the dialogue keeps naming them directly:
- no dictionary inside the model
- no grammar rules inside the model
- no separate reasoning unit or language module
- next-token prediction as the core loop
- knowledge "smeared" across layers instead of looked up from a database
- GPUs as the thing that makes the model "happen"
- context windows as the limit on continuity
According to Max Leiter's story, even interpretability shows up as a joke about mapped features for honesty and the Golden Gate Bridge. It is a pretty good creative trick, because the explanation arrives as dialogue instead of explainer prose.
The disclosure line became part of the story
Discussion around They’re made out of weights
Thread discussion highlights: - mjg2 on meta / author intent: "Weights helped me draft and proof this story." ... "there is no intelligence found in either the author nor their tool." - ACCount37 on thinking as computation: "There are exactly two possibilities: thinking can be expressed as computation, or thinking requires hypercomputation." - pona-a on LLM consciousness skepticism: "LLMs aren't; their subjective 'experience' restarts every session... claim this echo of humanity is itself conscious ... is either a product of psychosis or some long-standing hatred for life."
Leiter closes the post with a one-line note that weights helped draft and proof the story. In the HN discussion roundup, one of the earliest quoted reactions turns that into a meta argument about whether there is "intelligence found in either the author nor their tool."
That is the most interesting second-order effect here. A short adaptation about statistical language landed as a live test of how readers feel about AI-assisted writing when the assistance is disclosed plainly instead of hidden.
The thread drifted into consciousness fights
Fresh discussion on They’re made out of weights
Today’s new discussion mainly deepens the philosophy-of-mind angle rather than adding technical detail. One fresh comment argues that the real fork is whether thinking is computable at all, rejecting “hypercomputation” as magical thinking and treating the question as whether minds are just math implemented in matter. Another fresh comment pushes back harder, saying LLMs are only useful tools, that their subjective experience is not continuous across sessions, and that calling them conscious is a serious category error. So the new signal today is less about the story itself and more about whether “weights” implies anything like genuine cognition or consciousness.
By the next day, fresh HN discussion says the new signal was less about the story's craft than the old argument underneath it: whether thinking is computational at all, and whether LLMs should be treated as anything more than useful tools.
The thread's cited fault lines were simple:
- one camp argued the real question is whether thought can be expressed as computation
- another argued session resets and tool-like behavior make consciousness claims a category error
That drift tells you what kind of object this became on Hacker News. It started as a clever adaptation and turned into a philosophy-of-mind honeypot.
Memory is the last twist
Summary: "They're Made Out of Weights" by Max Leiter
"They're Made Out of Weights" is a short story written by Max Leiter that uses a dialogue-driven format to demystify Large Language Models (LLMs). The narrative illustrates that LLMs operate entirely through complex mathematical weight multiplications across multiple layers, rather than relying on explicit grammar rules, dictionaries, or human-like consciousness. The story frames LLM reasoning and knowledge as emergent patterns derived from these weights, while noting the constraints imposed by context windows and GPU dependencies.
The final new beat in the story is not about weights at all. It is about memory. After all the talk about context ending and users being "just a dream" to the model, the closing exchange says the next generation ships with persistent memory across sessions, described as the company's most requested feature.
That ending changes the temperature of the joke. The story spends most of its runtime insisting that the machine forgets, then closes on the one product change that would make the relationship feel less disposable.