Stereotype Threat
Inference and prejudice
Universities need to be thinking harder about how weights and inference matter to being “AI ready.” No campus conversation I have heard is addressing the “sameness” problem in the context of LLMs. Universities cannot keep serving boilerplate. They cannot simply give faculty and students access to a mid-range stable model and call it a day while the frontier is moving. That would be like giving everyone decades-old textbooks.
Some context. Before it was a problem, a stereotype was a solution. How do you print a piece of work at scale? The answer was to cast a solid metal plate from a page of set movable type, called a stereotype, first invented in 1725 but not commercially viable until around 1800. A print compositor would set the type, then take a plaster mold, then pour molten type-metal to make one rigid plate of the whole page. The original type was taken apart and put back into the letter case to set the next page. The first stereotyped book printed in America was a Bible, in 1812, from plates the Philadelphia Bible Society imported from England. The American Bible Society, founded 1816, made stereotypes its business model. By the 1820s half of American-made Bibles were stereotyped.
A stereotype makes sense with stable works (like the Bible), where you’re not likely to change a single word, and you are expecting a big print run. You can sell the print run and you can also rent or sell the plates to other printers and publishers, who can re-use them, but can’t change a word. This matters for my field (literature) because publishers decide early what to stereotype (like Uncle Tom’s Cabin) and what not to (books you’ve never heard of). Famously, Ticknor & Fields, the publishers of The Scarlet Letter (1850), did not expect it to sell quickly and did not stereotype the first run. They scrambled and set an error-filled second run, and then finally cast stereotype plates for a third edition.
Once you make a plate it isn’t useful for anything else except printing that page of text with ink and a fresh piece of paper. Newspapers, which cast fresh plates daily, melted them down at the end of the day (unless it was an important day). Book authors sometimes asked for their plates when a publisher was done printing (and the book wasn’t going to be a runaway best seller).
Today you read headlines about hundreds of billions of dollars being spent a year on inference chips, processors that do one thing, usually running a trained large language model (LLM). Google is reportedly developing a chip, nicely called Frozen v2, that would permanently etch Gemini’s computing logic onto the silicon wafer, “akin to ‘freezing’ the software in the chip.” Is it a leap to compare an inference chip to a stereotype? No, in fact. The inference chip holds the finished model as a mass of frozen numbers, called weights.
AI models read and write in “tokens,” which follow similar principles to a type case, which was weighted toward which letters are most commonly used. A type case holds far more e’s than z’s, more vowels than consonants. You need letters in the proportions they appear. When movable type was set by hand, the most common letters were close at hand. Frequent pairs and words came pre-cast as single pieces — fi, fl, and the “&.” A token set is built the same way: a large corpus is scanned, the character sequences that appear together most often are merged into single units, common whole words become one token (like “and” was one sort), and rare words are assembled from smaller pieces. The model then generates one token at a time, predicting each from the likelihood of what follows what.
An LLM is one enormous formula for predicting the next token, and the weights are its settings: hundreds of billions of numbers whose values, in combination, determine what the model writes. A model is trained by scanning a corpus that includes just about everything written often enough to survive, which is why so many stereotyped texts are represented in the corpus.
Training adjusts the model’s prediction toward what texts in the corpus tend to say. When training stops, the weights are fixed, somewhat similar to the way a page of type is fixed before the molten metal is poured to make a stereotype. The model reads the fixed numbers from memory and uses them to process a response to a prompt, without a mechanism to change a weight. With a stereotype, every fresh sheet rolled against it looks the same. With an LLM, the answers vary with the prompt, but the same assumptions (weights) are operating. Everyone will get a similar answer to a similar question.
Some differences. First, the weights are a file, and a file can be copied exactly and endlessly, so millions of users can use it. Second, in this analogy the weights are the stereotype plate and the chip is the press. When a company retrains its model, it’s a new stereotype plate. Neither the stereotype plate nor the chip is “corrected” by editing at the margins. The Bible Society’s “Directions for Repairing Plates” describes physically working the metal, like an auto shop getting rid of a dent. If there is an error in a trained LLM, whole new weights are produced.
In printing, the expensive part is setting the type for the first plate. After that, each copy costs only ink, paper, and press time. There is the cost of the metal to make the stereotype plate and the craft of making the plate, but once that is done it is cheap to strike from it. Selling print copies from a stereotype plate is a stable revenue stream.
A data center is a factory that manufactures tokens, as Jensen Huang put it. Each token, he says, is a unit of profit, and he predicts high-end tokens will sell for up to $1,000 per million. The factories run two main chip designs: NVIDIA GPUs and Google TPUs, which keep the weights in high-bandwidth memory stacked next to the processor and stream them in; and Groq’s LPU, which holds the weights in SRAM spread across the die and uses no HBM. Groq states its SRAM is about 100 times faster than the HBM that GPUs use. Groq sells the output as “Tokens-as-a-Service,” billed per token.
The business decision to invest in upfront costs and then sell copies is the business decision to invest in inference. Training a frontier model (the ones you see in the news with the most cutting-edge ability) is the expensive part involving labor, thousands of specialized chips, hundreds of millions of dollars for the largest models. Once a model is trained, the output is cheap. You pay for each answer, at a rate per million. The model reads your input in tokens and writes its answer in tokens. You pay for both. The AI industry calculates tokens per second and tokens per watt.
Given the upfront costs, only producers with enough capital to buy a press and a set of movable type, and with a paying readership lined up, could go into the stereotype business. Frontier-model training is limited to a few firms, like OpenAI, Anthropic, and Google, that could bear costs before inference revenue. The model stays on their servers, your question travels to their data center through an account (called an API, an arrangement for sending work to someone else’s machine and getting the result back), their copy of the model answers on their chips, and you are billed for the tokens. You never touch the model.
The second way to make money is to hand out the stereotype plate to anyone who wants it. For years Meta gave away the weights of its Llama models: anyone could download the file of numbers and print from it on their own hardware, paying Meta nothing per copy. Meta has since gone closed. The Chinese company DeepSeek still gives its models away, and its January 2025 release matched the expensive American ones. These are called open-weight models. The downloader basically gets the stereotype plate without the manuscript: the weights are public but the training data, the reasoning why it works as it does, stays private. In the open weight paradigm the user can “melt it down and recast it” or in this case, train it further on new texts to produce new weights.
Why give away a stereotype plate that cost a fortune? Because the business model undercuts everyone selling impressions. In the print news era, syndicates mailed ready-made stereotype plates of articles and filler to small-town papers, which printed them as their own. This is the origin of the word “boilerplate.” Handing out plates spreads the output without spreading the ability to set a page, which would be competition. Training stays where the capital is.
My point in making this analogy (of course there are differences) is to reach people in the higher education sector who aren’t really thinking about AI in the right way. Higher ed is already tying itself in knots arguing about political polarization and anthropology and “character formation” and the despair of learning outcomes and whether anyone learns anything in college and why go. My view is that in the AI era, universities should be sites that focus more on frontier knowledge and unknowns (as I’ve written about re: the Rumsfeld Matrix). Teaching “known knowns,” which is what general education is, is less and less important in the AI era.
Both stereotypes and inference chips are an investment in stable knowledge and known knowns. An investment in inference assumes stable recorded knowledge. This is good, for many reasons. Stereotype plates put a Bible in every household. Inference puts competent answers within anyone’s reach for a fraction of a cent.
Frontier knowledge, outside the stable, fixed knowledge, remains expensive. Universities, I have been arguing, should be where new questions are asked and new knowledge is made. You hear the frontier AI model laboratories claiming their newest models are beginning to produce results that were in no corpus (an OpenAI model disproved a 1946 Erdős conjecture this spring and just yesterday Fable disproved the Jacobian conjecture). But their capital is amortized against serving answers at volume, which are overwhelmingly known knowns. The business case is still stereotype plates.
Remember textbooks? They were a stereotype plate business model which is why if you got two of the biggest states (Texas and California) to sign off on something you could make books that you could print by the hundreds of thousands if not millions. Now the token does it cheaper.
The four largest technology companies plan roughly $725 billion in capital spending for 2026, most of it on inference. The financing has raised eyebrows. The companies depreciate the hardware over five or six years while the life of the chips is probably two or three. Everybody’s funding each other: Microsoft invests in OpenAI, OpenAI spends the money on Azure, and that spending justifies Microsoft’s next round of capital. Demand looks independent but it isn’t. And there has been one market correction already: in January 2025, when DeepSeek’s free plates matched the expensive ones, Nvidia lost roughly $590 billion of market value in a single day, the largest one-day loss any company had recorded.
As the price per token continues to fall (and it will fall, since anyone can create a free cast at the cost of electricity; Dean W. Ball makes the same point from the policy side) print runs must keep lengthening for the stereotype to pay. The business case requires canonical, stable works like the Bible. The UK’s AI Security Institute measures open-weight models trailing the newest proprietary models by four to seven months on its cybersecurity evaluations, down from six to ten months a year ago.
Will consumers just keep wanting consensus, known known “answers” from chatbots? I hear complaints already about sameness and people coming to the same ideas at the same time. A paper in Trends in Cognitive Sciences this March argues that when different users’ expression runs through the same models, their distinct styles, perspectives, and reasoning strategies become homogenized. A study of 6,875 student essays found that quality improved with AI assistance while structural diversity collapsed, the essays losing 70 to 78 percent of the variance in how they were built. A Georgetown neuroscientist analyzing college admissions essays before and after 2022 found that “suddenly everyone’s life got a lot more similar.”
The business case for inference is hard to argue with: Google processed 9.7 trillion tokens a month in May 2024 and 3.2 quadrillion a month in May 2026. But I keep reading that much of the growth is the LLM’s own verbosity. Newer reasoning models write long internal chains before answering, many times the tokens per request of their predecessors, and agent systems turn one task into dozens of model calls. It’s hard to tell whether the increase in tokens processed is a matter of new users or a chattier machine. I regularly tell Claude not to be so garrulous.
Textbooks got old quickly. It was an expensive technology so people kept using out of date textbooks. Similarly, released weights cannot be recalled. Copies sit on computers everywhere, and will continue spitting out answers to prompts.
As a scholar of the nineteenth century, I’m still irritated at Walter Lippmann for hijacking the term “stereotype” in 1922 to mean a fixed, reductive mental image of a group. Lippmann’s stereotype is a judgment cast before the particular case arrives. That is what an inference chip is, in silicon. Judgment is fixed in training, months before you type your question, and the chip applies it to each fresh case. There’s a reason the industry calls this inference, a conclusion drawn from premises that are already set. (Little known fact: “cliché” is the French term for a stereotype block, from the sound of the matrix striking the metal.)
Will people keep paying for impressions struck from the same cast, and for how long? If the models have started settling old math conjectures, who recasts the plates, how often, and who pays? And what is a university for when the known knowns cost a fraction of a cent. Is the point to keep selling impressions from a stereotype in the classroom? Is this conversation even on the table at most universities?
I start with a few immediate references. First, I was waiting for the other term to fall: "cliché," an onomatopoetic metonymy that records, allegedly, the sound of the stereotype plate hitting the lead. And second, Roland Barthes' conception of the readerly in his S/Z as the already-seen, already-read, the déjà-vu and déjà-lu. He was already thinking about these problems!
1. The printing analogy may conceal as much as it reveals.
A stereotype plate stores a determinate page. Model weights, if I understand this correctly, do not store a determinate answer; rather, they encode distributed dispositions that produce different outputs depending on prompts, context, sampling, system instructions, retrieval and later tuning. You acknowledge variation but continue to speak as though inference were fundamentally the repetition of an already composed impression: “judgment is fixed in training” and then simply “applied” to each new case. That risks confusing constraint with predetermination. The consequential question is not merely whether the weights are fixed, but where novelty, adaptation and correction can enter a system whose weights remain fixed. Your analogy may be excellent for the economics of replication while being much less adequate for the epistemology of generation. How then to think about even the known knowns as enigmatic, that is, as capable of surprise?
2. The argument against teaching “known knowns” does not follow from the economics of cheap inference.
Even granting that established information has become extraordinarily cheap to retrieve, it does not follow that universities should shift away from teaching it. Knowledge is not educationally valuable only because it is scarce. Students need internalized facts, procedures, histories and conceptual distinctions in order to recognize bad and worse answers, formulate worthwhile questions and understand frontier research. Perhaps you move too quickly from “machines can supply competent answers cheaply” to “general education is less important.” In fact, cheap and abundant answers may make disciplined foundational knowledge more important, because judgment cannot be exercised entirely through outsourced retrieval. The university’s alternative to boilerplate—love this word!—is therefore not simply “the unknown”; it is also the cultivated ability to distinguish established knowledge from plausible-sounding repetition.
Perhaps, then, “known knowns” need to be divided. Some are good known knowns: hard-won, collectively tested facts and practices that become foundational not because they are permanently fixed, but because we know how they were established and under what conditions they might fail. Latour calls these black boxes: we don't need to open them—so long as they work. But other known knowns are bad: propositions and associations that appear settled only because they have been repeated, naturalized or statistically overrepresented—the stereotype as epistemic matrix, the cliché as its recognizable surface. One danger of AI is that it can make the latter look like the former, delivering inherited regularities and reliable knowledge with the same fluency and confidence. When you're AI feels like a mirror, worry. A lot.
And then there is Žižek’s missing category, which we've discussed: the unknown knowns: the things we do not know that we know, the disavowed assumptions and tacit routines that organize inquiry before a question has even been asked. Here AI may be less the university’s external antagonist than its disturbing double. It makes conspicuous how much academic work already consists of recombination, credentialed paraphrase, managed novelty and, um, boilerplate. The university’s task, then, cannot simply be to leave the known knowns to machines and move triumphantly toward the frontier. It must test, preserve, and teach the good known knowns, while reopening the bad ones to scrutiny, bringing its unknown knowns into view, which of course, can't ever be completed. That is also how the known might become enigmatic—and capable of surprise—again.
The discovery of the Jacobian counterexample for N=3 with the LLM approach, after over a century of determined efforts by thousands of genuine mathematical geniuses, and at what is still fairly early on in this revolution is beyond astounding. That subsequent to acheivements of this sort there are still people who downplay AI capability not just now but as a prediction into the medium term future is just a testament to human stubbornness and imperviousness to the weight of evidence accumulating right before their eyes.