• Bezos's Insight: In 1994, Jeff Bezos recognized that the internet provides free shelf space for millions of books, revolutionizing retail by enabling infinite listings unlike physical stores limited to about 100,000 titles.
• Infinite Shelf Limitation: Infinite shelf space requires infinite inventory; Amazon succeeded due to ongoing book production by publishers and authors, maintaining supply.
• LLM Potential: Large language models (LLMs) can generate infinite text permutations, solving inventory issues by creating content on demand from probability distributions.
• Current Interaction Flaw: Interaction with LLMs occurs via a basic text box, likened to shouting prompts at an oracle, lacking structured browsing like aisles in a store.
• Image Generation Analogy: Generating all possible 100x100 black-and-white images yields vast static, illustrating that infinite possibilities exist but navigation through such spaces is impractical without structure.
• LLM Solution to Text Space: LLMs compress possible text into navigable distributions, generating plausible books without exhaustive search, avoiding the inefficiencies of full enumeration.
• Latent Library Concept: Latent Library is an interface where LLMs hallucinate categories, titles, and books dynamically as users browse, with citations linking to newly generated content, building explorable paths.
• Hallucinations as Inventory: The project reframes LLM hallucinations as deliberate features for fiction-like generation, providing a discovery-based interface over direct prompting for exploring potential texts.

---

As the story goes, Jeff Bezos is sitting somewhere in 1994, probably in a garage or a warehouse or wherever people had ideas before they had LLMs, and he realizes something important about the internet.

Shelf space is now free.

Not cheap. Not affordable. Free.

In a physical bookstore, every book costs you square footage. Real estate. Prime placement at eye level versus that bottom shelf where dreams go to die. You can stock maybe 100,000 titles if you’re Barnes & Noble and you’re trying really hard.

On the internet? You can list millions of books. Every book that exists. Every book that’s ever existed. The Long Tail. The infinite shelf.

This, Bezos correctly understood, changes everything about retail.

Wall Street analysts never bring up the main problem with this idea, though. Infinite shelf space is really only useful if you have infinite inventory.

[

](https://substackcdn.com/image/fetch/$s_!ltDr!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F65cc28ed-1d09-4bdc-8490-34a59e9bee6b_1584x876.png)

Amazon worked because publishers kept making books. Authors kept writing. The supply side was healthy. Slow, human-paced, occasionally stopping for lunch or letting deadlines go whooshing by, but healthy enough that the shelves never looked too empty.

But what if you could have infinite inventory though?

What if the shelf space and the books could both be infinite?

The Oracle and the Straw

---

Rohit Krishnan started a recent post off with a couple sentences that really put into words something that’s been on my mind for a while:

“Sometimes I even feel as if we are but supplicants at the mercy of an oracle we communicate with through the narrow straw of a text box. Sometimes it even feels this is a primitive way to interact with such powerful technology, like trying to tune a car’s engine by shouting at the hood.”

[

Strange Loop Canon

Poisoned prose

“make no mistake: what we are dealing with is a real and mysterious creature, not a simple and predictable machine” Jack Clark, Cofounder of Anthropic…

Read more

19 days ago · 28 likes · 18 comments · Rohit Krishnan

](https://www.strangeloopcanon.com/p/poisoned-prose?utm_source=substack&utm_campaign=post_embed&utm_medium=web)

We have these models now that can generate essentially any permutation of tokens up to their output limit.

Every possible arrangement of words that could exist, we can now sample from.

And we interact with them through… a text box.

Type a prompt. Get response. Type better prompt. Get better response.

It’s like Bezos built the infinite shelf but forgot to build aisles. Or a front door. Or any way to browse that isn’t just shouting the ISBN at a warehouse robot.

(The warehouse robot is very patient but also very literal.)

A Confession About Static

---

Back in undergrad (and I’m telling you this because it’s relevant, or maybe because I still haven’t stopped thinking about it…) I had this realization during an image manipulation project.

Any image you can possibly imagine is just a specific arrangement of pixel values.

Which means you could, theoretically, write a program to iterate through every possible combination of pixels. Eventually, you’d generate the Mona Lisa. A photo of me with my future grandchildren. A perfect snapshot of tomorrow’s sunset. Every frame of a movie that hasn’t been made yet.

I tried it with a 100x100 black and white image.

My program showed me static.

So much static.

The math is simple: 2^(100x100) = 2^10000 ≈ 10^3010 possible images.

For context, there are about 10^80 atoms in the observable universe.

My little program was trying to search through a space larger than reality itself.

Every image exists somewhere in that space. Every masterpiece is just a number away.

The problem was navigation.

The problem was that “infinite possibility” and “useful possibility” are not the same thing.

[

](https://substackcdn.com/image/fetch/$s_!yHyo!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb3ce6778-0204-4b87-b72d-f7abb5e8de77_100x100.png)

One of 10^3010 possible images

Fast Forward to Now

---

Large language models have solved a version of this problem.

Not exactly searching through all possible text sequences, which would take longer than the heat death of the universe and require more computational power than exists. That space is just as impossibly large as my pixel problem. Larger, even, if you’re keeping score.

They solved it by learning the structure of useful text. By building an intuition for what tokens should follow other tokens. By compressing the library of all possible books into a probability distribution we can actually sample from without waiting for the heat death of anything.

You can ask an LLM for a book about competitive vegetable gardening, and it won’t iterate through 10^billion possible books showing you static. It’ll generate something plausible. Something readable. Something that feels like it could be a real book from a timeline slightly to the left of ours where people take turnips much more seriously.

The infinite shelf problem is solved.

We can generate infinite books.

We just don’t have a great way to browse them yet.

We don’t have the aisles and the categories and the “customers who liked this also liked” recommendations.

We built the warehouse and then stopped.

Enter: The Latent Library (Where Hallucinations Are The Whole Point)

---

So I built this thing. Or Claude built it. Or “we’re’“ building it. The distinction gets fuzzy with these projects.

It’s called Latent Library.

Every category is hallucinated by an LLM.
Every book is hallucinated by an LLM.
Every citation in every book leads to another hallucinated book.

Eventually we’re going to have different ways for you to collaborate on generating the contents of the books.

Nothing exists until you look at it.

Everything exists the moment you look at it.

It’s the solution to Bezos’s infinite shelf problem, but the books materialize as you browse, like Schrödinger opened a bookstore and forgot to tell the books whether they exist or not.

You click on a category like “Programming Fables” and see titles:

• The Fox Who Compiled Himself

• Rubber Duck Debugging: A Memoir

• The Tao of Semicolons

• grep Dreams of Electric Strings

None of these books exist.

You click on one. Now it exists. Soon, we can start generating chapters. Those chapters can have citations to other books that don’t exist yet. You follow a citation. That book begins existing too.

The library grows in the directions people actually explore, like a city that builds streets only where people walk.

Hallucinations as Features, Not Bugs

---

We keep apologizing for hallucinations.

“Sorry, the model hallucinated that fact.”
”Disclaimer: This may contain hallucinated content.”
”Please verify, the AI sometimes makes things up.”

But what even is a hallucination in the context of fiction though?

If I ask an LLM to generate a book called “The Recursive Raven” about a bird stuck in an infinite loop, and it generates a plausible-sounding book… is that a hallucination? Or is that just generation?

The LLM isn’t claiming the book exists at Barnes & Noble. It’s not confusing itself about reality. It’s generating a book that could exist. A book from the space of all possible books.

That’s not a bug.

Latent Library embraces this. Makes it central. Turns it into the core mechanic.

Browse → Discover → Materialize → Explore → Discover → Materialize.

Hallucinations aren’t failures. They’re the inventory.

They’re what you’re here for.

The Interface Is the Insight

---

I know what you might be saying. “LLMs generating text? Solved problem”. And I agree, that’s old news, especially in AI years. What I want to call your attention to is the interface pattern. That’s what I think is interesting.

Instead of this:

USER: “Generate a book about quantum mechanics for cats”
LLM: [50,000 tokens about Schrödinger’s litter box]
USER: [reads or doesn’t, probably doesn’t]

We have this:

USER: [browses “Physics for Felines” category]
USER: [sees “The Cat’s Guide to Superposition”]
USER: [clicks]
BOOK: [begins existing]
USER: [reads Chapter 3: “Nine Lives, Infinite States”]
CHAPTER: [references “Advanced Paw-ticle Theory”]
USER: [follows citation]
NEW BOOK: [begins existing]

It feels like discovery instead of generation.

Maybe at some level there is no difference.

Maybe discovery is just generation with better PR.

All Books Already Exist, They’re Just Waiting to Be Read

---

[

](https://substackcdn.com/image/fetch/$s_!FKLO!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8093e10a-8bb6-460c-848e-a2a592946a27_1232x928.png)

Borges wrote about the Library of Babel. A library containing every possible 410-page book made from a specific set of characters.

Most books are gibberish: “fhjksd jsdfkljsd fsdkljfsd...”

Some are mostly gibberish with occasional words: “fhjksd the jsdfkljsd cat fsdkljfsd...”

A tiny fraction are readable.

An impossibly smaller fraction are worth reading.

Somewhere in there, theoretically, is the perfect book. The one you needed. The one that answers questions you didn’t know to ask. The one that makes everything suddenly make sense at 3 AM when you’re supposed to be sleeping but instead you’re thinking about bookshelves.

LLMs are like having a librarian who’s memorized large sections of that library. Enough to say “oh, you’d like something like this” and pull out something worth reading without making you wade through the gibberish sections first.

We just needed a better way to browse than shouting into the void and hoping the void shouts back something useful.

Back To “The Thing”

---

Latent Library. It’s a browsable interface for the space of possible text. It treats generation as exploration. It makes hallucination the feature instead of the bug. It lets you wander through categories and citations and reviews until you find something you didn’t know you were looking for, which is how all the best discoveries happen anyway.

Will anyone use it? Unknown.

Is it useful? Depends what you mean by useful.

Is it interesting? I think so. But I also thought iterating through all possible pixel values was interesting, so up to you on whether to trust my judgment.

We solved the infinite inventory problem. We can generate anything. We just haven’t built enough interesting ways to navigate that infinity. We’re still shouting at the warehouse robot when we could be browsing the aisles.

Text boxes are fine. Prompts are fine. They work.

But maybe we need more interfaces that feel like discovery instead of creation though.

Maybe we need more ways to browse the space of possible outputs.

Maybe infinite possibility is only useful if you can navigate it.

Try It Yourself

---

Latent Library is available at https://latentlibrary.xyz.

Browse around. Find a book. Pretty soon you’ll be able to read it too.

In one sense, it didn’t exist before you found it.

But in another, it always did, waiting patiently in the space of all possible books in Borges’ Library of Babel, like a good idea waiting for someone to have it.

All books exist in potential.

You get to choose which ones to collapse into reality.

Share

---

If you do decide to browse the library, you might notice one author, Elara Voss, being a bit more prolific than the others. It turns out this author and name is likely a complete invention by LLMs. If you’re curious you can read more about that here:

[

Read Max

Who is Elara Voss?

Greetings from Read Max! In todays’s newsletter, two items…

Read more

3 months ago · 200 likes · 50 comments · Max Read

](https://maxread.substack.com/p/who-is-elara-voss?utm_source=substack&utm_campaign=post_embed&utm_medium=web)

Harvard has finally acknowledged that it has a problem with grade inflation. Unfortunately, though the university has now admitted that its instructors award too many A’s, its proposed solutions fall short.

According to a 25-page report authored by Dean of Undergraduate Education Amanda Claybaugh and released to faculty and students last month, Harvard now gives top marks more than 60 percent of the time, up from just 25 percent in 2005. The median GPA has also increased, to 3.83 from 3.49 in 2005. According to the report, Harvard is “failing to perform the key functions of grading,” which is “damaging the academic culture of the College.”

The report marks a sharp shift in tone for Harvard. A 2023 report cautiously cited compression—that is, a tighter clustering of grades but not necessarily a higher average—but Claybaugh’s report explicitly identifies grade inflation. It also outlines several incentives—some driven by the university, others by students and faculty—fueling the problem.

Though it doesn’t explicitly say as much, Claybaugh’s report also reveals the consequences of Harvard’s waning commitment to merit. She notes that over the past decade, faculty have relaxed grading standards partly in response to the university’s “exhorting faculty to remember that some students arrive less prepared for college than others.” Harvard attributes this unpreparedness to students’ limited exposure to complex texts and shorter attention spans caused by “media culture.” But during this same period, the university made standardized tests optional, sidelining one of the few objective measures of academic readiness in admissions. (The report neglects to note this.)

As courses became less demanding and grading more lenient, the report explains, earning an A grew progressively easier. Harvard’s shift from high-stakes exams to frequent, lower-stakes assignments rewarded effort over mastery, giving students a false sense of confidence in their understanding of a subject. Some faculty even gave students A’s just for completing assignments.

Part of the pressure to give A’s comes directly from students. When nearly everyone seems to be receiving A’s, even the possibility of an A- seems terrifying to students. Faculty worry about upsetting students and doubt that administrators will “have their back” if the student complains. Students routinely negotiate grades with professors, while professors, especially non-tenured ones, fear receiving negative student evaluations.

Even tenured faculty are not immune from student pressure. Poor course evaluations and a reputation for harsh grading can lead to smaller class enrollments, which affect department budgets. As one faculty member put it, giving honest grades now requires great “strength of character.”

Harvard’s approach to solving grade inflation appears incremental and tentative, though the problem is perennial. Most of the university’s proposals accept grade inflation as a given rather than tackling its underlying causes. For instance, Harvard suggests expanding the grading scale to include A+, which would be awarded only to a limited number of students in a class. But this merely kicks the can down the road, much like raising the debt ceiling does nothing to tackle the problem of spending growth.

Another proposal would introduce a “variance-based grading system” for internal use, allowing Harvard to distinguish high-performing students for awards and honors without publicly confronting lenient departments or underperforming students. That’s just a band aid that doesn’t correct relaxed standards.

Some of the more promising proposed reforms include eliminating student course evaluations to ease pressure on faculty, recording median grades for each course on transcripts to discourage students from selecting only easy courses, and returning grade distributions to 2015 levels—a time “when grading was not overly stringent . . . but we did assign a broader distribution of grades,” according to the report. These would all represent changes in the right direction, albeit still modest ones, suggesting something less than a commitment to academic rigor.

If Harvard really wants to address grade inflation, it needs a reset. One solution would be to institute a standardized, university-wide grading curve. In the past, when grade quotas or semi-curves were suggested, Harvard officials insisted that “nobody wants to legislate grade distributions.” But the situation is far more serious today. A temporary curve lasting two or three years would not only allow faculty to apply strict grading standards guilt-free but also help modify student expectations.

By tolerating grade inflation, Harvard and its peers are misleading students about what excellence really entails and denying young people the elite education they have promised to deliver.

Photo by Aaron M. Sprecher/Getty Images

Margaret Atwood has a reputation for spookily accurate predictions about the future of humanity in her novels. This is quite strange, because — as her new memoir Book of Lives demonstrates — she is in fact quite obtuse when it comes to interpreting humans. She doesn’t clock male romantic interest, even when obvious; is blindsided by rivalrous feuding in an independent publishing house of all places; can’t work out why so many “good-looking, well-dressed single women” would be sitting alone at Prague bars in 1984, and has to be told. When her father is gravely ill, she doesn’t “quite understand” why her mother goes to the hospital daily to play him his favourite Beethoven recordings: “Surely a couple of times a week would do.” Later, as her own partner lies dying, she finally gets the point.

She peppers her memoir with “why” questions, as if anxiously monitoring for things she might have missed. She also poses puzzled questions of her past self, though seems refreshingly untroubled by the answers. Of her temperament, she explains, “I was more interested in the paper people I could create than in doing deep dives into my own psyche, should such a thing exist”.

Being good at myth-making, the still fizzy 85-year-old seamlessly weaves her interpretative deficits into a vivid origin story: “One of my theories about novel writers is that they don’t know more about human nature than other people: they know less, and their novels are attempts to figure it out.” Her father was an entomologist; like him, she can’t stop mentally cataloguing bits of the world, then cross-referencing. “If I watched carefully, I’d be able to find out how stuff worked, whatever that stuff might be. Sometimes it was mechanical devices, such as sewing machines. But usually it was people.”

But though minds were slippery things to master, in other informational domains Atwood was in her element. She relished learning about arcane codes and ciphers, then applying them. She became able to identify rare fauna and flora, read palms, do star charts, detect ghosts. For a time she did graduate work in English literature at Harvard, but the role of critically minded theoretician didn’t seem to fit; her interest in the world is much more practical. She loves building novels out of images and ideas, just as she also likes making dresses, cakes, puppet shows, comic strips, operettas, poems and jokes.

Though it isn’t clear where all that creativity came from, the practical bent appears to have been non-negotiable during an extremely adventurous childhood. Book of Lives cheerfully describes a dangerous existence in the company of intrepid parents and an older brother: living in remote parts of Quebec and Ontario, sleeping in tents, surviving near-drownings, casually seeing off snakes and bears, and always feeling cold. Any reader using a screen to bring up their child will presumably hang their head in shame.

From her stoic Nova Scotian forebears, “Peggy” (her name at home) also inherited a now-unfashionable dislike of navel-gazing and self-pity. Her relatives would “think it bad manners to show off, to snivel and complain, or to express emotions to excess, or even at all”. Later on in life, she shows impatience with the depression of her husband, self-satirising as “Ms. Fixit”. “Are you feeling better now? How about now? Look, we have a fondue pot! Doesn’t that make you happy?”. The marriage doesn’t last.

It is fun to cross-reference these personal insights with her novels. Like their author, the stories are madly creative; unsentimental and often macabre; crammed with fascinating empirical detail; fond of cross-referencing between domains. They are also short on psychological analysis, numinously infusing the natural world and manmade objects with unconscious feeling instead. Her most famous natural object is the female body: imagined as food (The Edible Woman); as decaying medical material (Bodily Harm), as breeding stock (The Handmaid’s Tale). The theme has made her enormously popular with feminists, a fact which she plainly experiences as a mixed blessing.

Certainly, Atwood is not an obvious candidate for leading the women’s movement. She loves chumming around with the male sex, and it takes until the account of her university days in her memoir to get a proper description of a female friendship. She has no truck with victimhood — the perpetrator of a grim sexual assault at graduate school gets a quick, heartfelt curse before the narrative moves on. And when, in the Seventies, a film director demands a “naked woman wrapped in cellophane” in a screenplay she is writing, she happily complies.

She is also markedly ambivalent about other women. Though she may not understand much about female psychology in detail, she is at least well aware of its toxic side. In Cat’s Eye, Atwood forensically detailed the brutality of nine-year-old girls to each other. In the memoir, we get the real-life version, with young Peggy as the hapless victim. She also knows from experience how envious and vindictive grown women can be. At one point, the protagonist of the dystopian The Handmaid’s Tale grimly accuses her absent mother, a second-wave feminist: “You wanted a women’s culture. Well, now there is one.”

Her political outlook is mostly liberal, which puts her at odds with the prevailing mood in modern progressivism — often to her credit. A big fan of freedom of expression, she introduced Salman Rushdie on stage shortly after the fatwa. She also signed the Harper’s letter against an “intolerant climate” in 2020. And she is very keen on due process, having no truck with the score-settling excesses of the #MeToo movement.

During this time, Atwood defended a University of British Columbia professor from what seems to have been a vicious witch-hunt. Afterwards, the witch-hunters tried to come for her as well, though she batted them off with her enormous fame, plus an acerbic essay entitled “Am I a bad feminist?”. In her memoir, she reflects that “when cults are at their height, fairness and human rights go out the window”, and is similarly critical of Trump’s lack of regard for legal process now.

Equally, though, she has been mostly silent on the authoritarian gender cult still operating in her Canadian backyard, except to disseminate some dodgy information about chromosomes and get annoyed when Hadley Freeman asked her about it. She has apparently reflected little about why chromosomally average teenage girls might be queuing up to get their sex organs surgically expunged. This has resulted in a lot of disappointed women shouting at Atwood on the internet. But there are suggestive clues to the blindspot in Book of Lives.

I would guess that Atwood can easily understand what’s wrong with the authoritarian coercion of people by outside forces — simply applying the mechanical model of “object x exerts force on object y”. But when it comes to minds that enthusiastically embrace damaging social norms inherited from the wider culture, it is quite possible she doesn’t really grasp the problem. Maybe this is because she is a classic liberal, uninterested in unverifiable claims about false consciousness; or maybe it’s because she struggles to place herself imaginatively within the mind of anyone sheep-like enough to succumb to social pressure — as she herself is manifestly not. And there is also the fact that in her work she often seems to exhibit profound unease about the female body. It’s part of what makes her artistic vision so compelling, even if it arguably compromises her politics.

I don’t really care about any of that; I like her books too much. But her memoir does rather undercut any flattering vision of The Handmaid’s Tale as a prescient forecast, peculiarly attuned to the authoritarian Right-wing future allegedly lying in wait for women, now that Trump is here. Atwood enjoys this image of herself as a far-sighted sage very much, referring to it several times. Perhaps it’s a small compensation for all those times her male pals laughed at her for not understanding something. But honestly, the idea that, in the near future, Western governments will need to use direct force to make women do market-friendly things against their own interests is now surely, definitively preposterous. The case of genderism — and surrogacy, and “sex work”, for that matter — shows that authorities only need to persuade enough women that certain activities are kind, or glamorous, or nobly self-improving; at which point, tender-hearted armies will rise up to ruthlessly punish dissenters themselves.

---

• Jarvis in Business: Fictional AI Jarvis from Marvel appears frequently in corporate contexts like boardrooms and executive talks to describe generative AI's potential.
• Nonthreatening Reference: Companies use Jarvis as an accessible, positive way to explain AI to public, press, and employees without technical jargon.
• Executive Endorsement: Akamai CTO Robert Blumofe frequently references Jarvis as the ideal for artificial general intelligence in presentations.
• Character Evolution: Introduced in 2008 Iron Man film, Jarvis evolves from household manager to suit designer and Stark Industries operator, inspired by comic butler Edwin Jarvis.
• Current AI Limitations: Today's AI agents perform narrow tasks with less reliability than Jarvis, focusing on specific industries or actions like buying groceries.
• Cultural Touchpoint: Jarvis provides a positive cultural reference for AI, contrasting negative depictions like Hal from 2001 or Terminator, and addresses anxieties about job replacement.
• Corporate Naming: Companies like Henry Schein One, Google (Project Jarvis, renamed Mariner), and former Jasper AI adopt Jarvis-inspired names for tools.
• Sci-Fi Influence: Tech draws from sci-fi like Star Trek for iPad, Jetsons for robots, and Star Wars for flying cars, with mutual influence between fiction and reality.

---

This is a scene that’s playing out all the time in businesses across America. 

Jarvis, the fictional AI from Marvel, was once known for helping billionaire playboy turned superhero Tony Stark save the world. But now he’s more ubiquitous in boardrooms, pitchdecks, corporate blogs and executive talking points.

---

Newsletter Sign-up

WSJ | CIO Journal

The Morning Download delivers daily insights and news on business technology from the CIO Journal team.

Subscribe

---

Why? Because the character, whose full name is “J.A.R.V.I.S.,” (standing for “Just Another Very Intelligent System,”) has become a nonthreatening, non-overly technical way for companies to describe the promise of generative AI in business to the public, the press and employees. 

Tech executives (who, frequently, are also nerds) love how accessible the reference is, but also how well it encapsulates the best of what AI could offer. 

“I use it a lot,” said Robert Blumofe, chief technology officer at Akamai Technologies. “If you think about [artificial general intelligence] and what we’re trying to do, I do think that Jarvis is sort of the best version of that that we have.”

AI Jarvis was introduced in the 2008 “Iron Man” movie that kicked off the Marvel Cinematic Universe, although he pays tribute to comic-book character Edwin Jarvis, the loyal (human) butler that worked for the Stark family and the Avengers. 

In the films, Jarvis’s capabilities expand from managing Stark’s household to helping him design his Iron Man suit, becoming something of an operating system inside and running parts of Stark Industries.

“He runs more of the business than anyone besides Pepper,” Stark said in 2015’s “Avengers: Age of Ultron,” comparing the AI to Stark Industries CEO Pepper Potts.

Fictional character Tony Stark as Iron Man in 2019's "Avengers: Endgame." Jarvis helps the billionaire playboy turned superhero manage his household, design his suit and act as an operating system inside it, and run parts of Stark Industries. Walt Disney/Everett Collection

Today’s AI agents, which can take some actions on behalf of humans, aren’t quite there yet. They are far less reliable and their abilities are typically more narrow, relating to particular tasks or particular industries, Blumofe said.  

But he loves the reference anyway, often using it in talks for both technical and nontechnical audiences. Society needs a cultural touchpoint to help explain the technical details of AI, but so much of what’s out there focuses on the worst of it, like the evil AIs from “2001: A Space Odyssey” or “Terminator.” 

“One of the things Jarvis does is he puts a really nice face on agentic systems,” said Julian Chambliss, an English professor at Michigan State University, and the Val Berryman Curator of History at the MSU Museum. 

Jarvis is moral and cognitively aware and dispels our anxieties that the technology could replace or even harm humans, said Chambliss, whose work explores the relationship between comics and culture. 

Vijoy Pandey had the same thought when his team at Outshift, Cisco’s innovation lab, would lovingly refer to some AI agents in development as “Jarvises.” There was even a running joke about making sure the appropriate guardrails were in place, “so that your Jarvis doesn’t turn into a Hal,” he said, referring to “2001’s” menacing superintelligence.

To be sure, some of today’s foremost AI experts think the Hal or “Terminator” scenario is more likely. In both instances, systems designed to serve humanity ended up turning against it. 

But the aspirational nature of Jarvis continues to inspire the growing cohort of sci-fi and fantasy fans running corporate America. Some companies are even formally naming their AI tools after him. 

SHARE YOUR THOUGHTS

Do you think Jarvis is an apt goal or comparison for today’s AI systems? What other fictional AIs should we be trying to emulate or avoid? Join the conversation below.

Dental technology company Henry Schein One has an analytics tool named after Jarvis. 

Google reportedly started building an AI agent that could navigate a web browser under the name “Project Jarvis,” although it was then reported that the company renamed it “Project Mariner.”  

And AI marketing company Jasper used to be named Jarvis. It rebranded in 2022 after trademark conflicts with Marvel, although the company said the new name better reflects the human-centered nature of its tools. 

Jarvis is far from the first time science fiction pop culture has collided with real world technology developments.

“There’s a long history in tech of leveraging movies as not only inspiration, but also to explain some of our goals,” said Sophia Velastegui, board director of BlackLine and former chief AI technology officer at Microsoft.

Velastegui, who also held product-focused roles at Apple between 2009 and 2014, said “Star Trek: The Next Generation” was part of the inspiration for the original iPad. 

New humanoid-style robots created to do household tasks have been compared with the Jetsons’ Rosey-the-robot. 

And flying-car racing company Airspeeder said it sees itself as delivering on the flying car promise of movies like “Star Wars: Episode I” and “Back to the Future.” 

At some point, Velastegui said, “It’s like, are we influencing the sci-fi movies or are the sci-fi movies influencing us?”

Chambliss agreed it is very much both. After all, he said, one of Iron Man’s computer assistants in the newer comics is called “Iron.GPT.”

Write to Isabelle Bousquette at isabelle.bousquette@wsj.com

• Mamdani's Voting Choice: Voted for himself on Working Families Party line, emphasizing loyalty to progressive movement over Democratic establishment.
• WFP Success: All 32 WFP-endorsed candidates won, including key positions like Public Advocate, District Attorneys, Borough Presidents, and City Council members.
• Progressive Influence: WFP and DSA represent rising power, extending to Albany for policy support like tax hikes.
• Establishment Shift: Progressives now form New York's political establishment, predicting decline of moderate elite.
• Election Dynamics: Mamdani secured narrow majority against scandal-plagued opponent Andrew Cuomo, who had low favorability.
• Campaign Tactics: Mamdani's victory driven by effective ground game and turnout of young voters, contrasting Cuomo's weak campaign.
• Progressive Strategy: Opposition to open primaries and election cycle changes preserved influence in Democratic primaries.
• Moderate Response: Centrists urged to build organizations or third party to challenge progressives, drawing from San Francisco model and New York's fusion voting history.

---

[

](https://substackcdn.com/image/fetch/$s_!GN3n!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F62d574dc-a189-4c5f-b84b-bdb5cf794631_6000x4000.jpeg)

Courtesy Michael M. Santiago/Getty Images.

When Zohran Mamdani, now mayor-elect of New York City, went to the polls last Tuesday, he didn’t vote for himself on the Democratic ballot line he had wrested from Andrew Cuomo five months earlier. Rather, Politico’s Jason Beeferman reported, Mamdani voted for himself on the line controlled by the Working Families Party, a third party that backs progressive challengers to Democrats in an effort to push them to the left. The vote was a nod to Mamdani’s outsider status—at the end of the day, his loyalty is to the union-backed movement the WFP represents, not to the Democratic establishment that preferred Cuomo.

But after Tuesday’s results, it’s not clear that Mamdani and his friends are still “outsiders.” Every single one of the 32 WFP-endorsed candidates won his or her election. These included not just Mamdani, but Public Advocate Jumaane Williams; the Manhattan and Brooklyn District Attorneys; the Manhattan, Bronx, and Brooklyn borough presidents; and 25 members of the 51-member City Council. As Joseph Burns wrote in City Journal following Mamdani’s primary win, the WFP’s influence extends to Albany, where Mamdani will need support to get his promised tax hikes approved. Mamdani’s candidacy also represents the rising power of another progressive org: the Democratic Socialists of America (DSA), the reach of which Mamdani aimed to grow when he first began running.

If anything, these two organizations, and the broader progressive movement, are New York City’s “establishment” now. Indeed, many are predicting the demise of the New York City moderate elite—some gloomily, some gleefully. Is New York fated, by the inalterable logic of history, to a future of “socialism on the Hudson”?

Not necessarily. Mamdani’s victory—and the broader progressive takeover of New York City’s government—isn’t really the result of a socialist revolution in public opinion. Rather, it represents a concerted political effort on the part of the progressives, combined with the complacency of the city’s moderates. The left, through groups like the DSA and WFP, staged an insurgency and won. If the center wants to retake the city, it needs to play the same game—and win.

One way to understand New York’s mayoral election is as a sweeping victory for the city’s progressive faction. But another is that their tribune secured barely a majority of the vote. This, in spite of the fact that his primary opponent had already lost the primary; was 20 points down in favorability in the days before the election; was implicated in multiple major scandals that had driven him from the state’s governorship; and is generally an object of hatred across the entirety of New York State politics. (Even worse: his other opponent was, horror of horrors, a Republican.)

As Jesse Arm recently wrote, the notable thing isn’t that Mamdani won, but rather how far he ran behind moderates and fellow executive candidates Abigail Spanberger and Mikie Sherrill in their respective races. With a different candidate running against Mamdani—a possibility discussed but never realized after the primary—it’s quite possible the race would have gone the other way.

[

](https://substackcdn.com/image/fetch/$s_!q8ST!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Feff90944-fbca-4d9e-9d02-c6ffd827a97e_1258x936.png)

Source: New York Times

Within his narrow margin, Mamdani’s success is attributable not so much to the inexorable march of progress as it is to his campaign’s superior tactics. After all, there is little doubt that if Cuomo had won the primary, a Mamdani challenge on the WFP line in the general would have been futile. But Mamdani’s primary success was at least in part about the effectiveness of his turnout machine, which drove tens of thousands of younger New Yorkers to vote for the first time. His ground game—powered by the DSA and WFP—was so effective that it transformed the entire primary electorate, resulting in a shocking polling miss.

Cuomo, by contrast, ran an anemic campaign, defined primarily by out-of-date ads and the occasional presser. Observers were left with the impression that the former Governor, scion of a political dynasty, felt entitled to his comeback. Talk to almost anyone in New York and they’ll tell you they spoke to Mamdani canvassers; no one remembers anyone in a “Hot Girls for Cuomo” shirt knocking on their door.

This inertia, though, is not limited to the mayoral race. Alvin Bragg, the arch-progressive prosecutor in Manhattan, barely faced a primary challenge; Eric Gonzalez, Brooklyn’s DA, didn’t even have to compete in a primary and was uncontested in the general. Infamous progressives on the Council like Tiffany Cabán, Sandy Nurse, Lincoln Restler, and Shanana Hanif faced no or no real primary challenge. All of these people will greenlight Mamdani’s agenda. In the case of radical changes like abolishing the NYPD’s gang database, they will probably be the ones pushing for change.

The progressives, moreover, know how important their ground game is to their success. That’s why the WFP worked to kill an amendment to the City Charter that would have created an open primary system—ensuring that registered WFP voters would be shut out so that they would have an easier time influencing the Democratic primary. And it’s why Mamdani voted no on proposition 6, which would have moved New York City’s local elections to the same cycle as national elections, increasing the size of the electorate and making organizing less effective. In other words, if you doubt that the progressive machine made the difference, recognize that the left sure thinks it did.

On election night, myself and fellow City Journal contributors

Rafael A. Mangual

and

John Ketcham

were joined in conversation by

Pirate Wires

’s Mike Solana. For me at least, the most interesting part of the conversation was Mike’s detailing of how San Francisco elected its current, more moderate leadership. In his telling, decades of dominance by progressives was undermined by concerted action on the part of the city’s more moderate residents, mobilizing around objections to progressive excesses, and focusing on the minutiae that are needed to win elections. To translate: moderates took San Francisco back by actually trying to win.

It would be easy for New York City centrists to throw their hands in the air and insist that Mamdani is the future, which we all had better get used to. But that would avoid the hard truth that this was a winnable race, and that Mamdani’s victory is the culmination of years of progressive insurgency largely unchallenged.

The progressive have won; they are the establishment. Which is why the moderates, rather than giving up, need to start acting like they’re the outsiders. If groups like the DSA and WFP are running the show, then moderates need their own institutions—even their own third party—to hold the new progressive establishment accountable.

The use of third parties for such purposes—buoyed by New York’s unique electoral “fusion” laws—is far from unprecedented. Bill Buckley’s quixotic 1965 campaign for the mayoralty was on a third party line; so was John Lindsay’s reelection bid four years later. Mario Cuomo narrowly lost to Ed Koch on the Liberal Party line in 1977; and Giuliani won on a Republican-Liberal fusion ticket in 1993, with the Liberal-line votes putting him over the top. Mike Bloomberg used the Independence Party line throughout his campaigns, both as a Republican and as a registered Independent. The WFP has not been the only party to try to discipline Democrats by selectively running candidates; the New York Conservative party did so as well, long after Buckley’s campaign was over.

A new moderate party could do the same for the numerous progressives now in charge of New York’s government. It could build enthusiasm for moderate candidates, taking advantage of the generous election laws that the WFP has used so successfully to its advantage. There’s at least one obvious electoral vehicle—the essentially moribund Liberal Party of New York, which endorsed Cuomo in 2025. If the LPNY can’t be wrested from its current controllers, then another party could be stood up easily—imagine a 2029 nominee running on the Prosperity Party line.

Even if a third party isn’t right, moderates need some sort of public-facing organization. The results last Tuesday indicated that a sizable fraction of New Yorkers are skeptical of their new Mayor. They can’t be welcomed into smoke-filled rooms, but they can join clubs, mailing lists, or any other infrastructure that might be on offer.

How could such a group hold New York’s progressive establishment accountable? San Francisco offers a good model: publicize and criticize the sort of excesses that scare off even New York’s left-leaning electorate. As recent MI polling showed, while Mamdani is popular in the city, many of his ideas are not. When progressives—not just Mamdani, but Councillors, borough presidents, and appointed officials—oppose fare enforcement or the city’s gifted and talented program, they’re at odds with a majority of Gothamites. Thus far, they’ve mostly managed to avoid scrutiny for these views; some adept social media wrangling could easily change that.

After all, if moderates are no longer the establishment, then the city’s problems are no longer their fault. There’s nothing quite so liberating as being out of power: you get to make all the criticisms, while your opponent has to take all the blame.

Which doesn’t mean that the city’s moderates need to be knee-jerk critics of everything Mamdani does. Doubtless, some of his initiatives will be popular with the electorate. Conservative talking points—no matter how correct—probably won’t persuade. That’s why the center-left needs to lead this movement—they’re the middle of the city.

But moderates can connect and have connected with New Yorkers before. There’s a reason why, in City Journal’s pre-election focus group, participants remembered the mayoralties of Mike Bloomberg and Rudy Giuliani more favorably than that of Bill de Blasio. The former provided safe streets and rising prosperity; the latter offered conflict with the cops and cronyism in city hall.

In four years, New York’s moderates might be mounting a successful campaign to stop a second Mamdani term. (I think their champion is obvious.) But if they do, it’s because they’ll understand that they haven’t failed to control just Gracie Mansion, but the whole of the city’s government. And they’ll try to retake that government from the ground up—after all, that’s how the left did it.

Share

The election of Zohran Mamdani as mayor of New York represents a seismic victory for the Democratic Socialists of America. Yet even as the DSA seems poised for a political breakthrough, it shows signs of unraveling.

Indeed, even before Mamdani’s victory last week, DSA members were intensely divided over whether his election would represent an electoral success or capitulation to the status quo. Some on the party’s radical fringes have argued against a consensus document meant to define DSA’s positions on Mamdani’s campaign. These tensions could shape Mamdani’s mayoralty and the future of socialist politics in America.

The DSA is not a political party. Like a party, however, the DSA has several factions and caucuses that jockey for power. Many of the DSA’s most prominent members belong to the Groundwork caucus, which emphasizes political reform and the need to preserve strong relationships with mainstream progressives and Democrats. These members are typically influential in the group’s politics.

But Groundwork often struggles to corral the DSA’s more radical wings. Groundwork representatives could not, for example, convince the rest of the DSA’s National Political Committee to endorse the No Kings Day protests.

These tensions mounted a week before the October 2025 meeting of New York City DSA, when the city chapter’s steering committee introduced two propositions: “Local Dues Drives Proposal” and “NYC-DSA Goals in Electing a Democratic Socialist Mayor.” While both documents provoked debate, conflicts flared in particular over one line in the second document: “If we succeed in electing Zohran Mamdani, our priorities will be campaigning for a democratic socialist mayor, expanding working-class power, and winning material improvements in the lives of the working class.” (NYC DSA did not respond to a request for comment for this article.)

Some members of NYC DSA and other chapters across the country feared this language would make Mamdani’s candidacy too central to the organization’s mission. Within days, five DSA factions—Springs of Revolution, Emerge Caucus, Libertarian Socialists, Reform and Revolution, and Marxist Unity Group (MUG)—published a flyer encouraging members to vote “No” on both propositions. They also released a petition demanding that the city chapter hold a “special meeting” in January with “true deliberative spaces, where members can collectively decide our work.”

There’s reason to suspect that one of those factions, MUG, has long been wary of Mamdani’s run for mayor. In August 2024, two months before Mamdani launched his campaign, someone writing under the name “Sid C.”—possibly Sidney Carlson White, a MUG member who now sits on DSA’s National Political Committee—penned a piece in the Socialist Tribune (a MUG-run Substack newsletter) arguing that it was “imperative that Zohran Mamdani . . . NOT run to win.” (a comment request to the newsletter’s email went unanswered.)

Instead, Sid C. argued, Mamdani should run an “agitational campaign” that advances the DSA’s positions on issues ranging from Palestine to police abolition in hopes of “lay[ing] the groundwork for our socialist future.”

Mamdani’s victory in the Democratic primary sent MUG back to the drawing board. In September, the entire faction published a missive that outlined its strategy for the “Mamdani moment.” The document demanded more “democratic” DSA institutions, with more “transparency and accountability” from the group’s leadership. It also outlined several principles about the DSA’s relationship to Mamdani, including asserting that “Zohran should be accountable to DSA’s democracy.”

The Liberation Caucus also opposed the pro-Mamdani propositions. Calling itself “Marxist-Leninist-Maoist,” the group formed this year to advocate for more revolutionary politics within the organization. Liberation Caucus recently issued its own response to the propositions, denouncing the NYC DSA for “seek[ing] to turn an ostensibly socialist organization into an electoral cheerleading and fundraising machine.”

“What worker would donate to NYC-DSA,” the group asked, “if all we do is run electoral campaigns?”

Remarkably, some of the Liberation Caucus’s members believe that Mamdani has become too conservative and friendly to Israel. One member called Mamdani “ZIOhran” and accused him of “pivot[ing] further and further to the right.” Another deemed Mamdani a “traitor and an opportunist,” adding, “[I w]ould not vote for him if I [were] a New Yorker.”

While some in the DSA’s radical fringes think that Mamdani is too mainstream, others see the mayor-elect as a revolutionary. In Partisan, a Communist magazine, someone writing under the name “Landry L.” argued that Mamdani’s mayoral campaign gave socialists an opportunity to “unite new and existing fights under a common banner.”

Mamdani’s campaign presents “a golden chance . . . to build power both inside and outside the state,” Landry L. writes. Invoking historian Noel Ignatiev—known for his calls to abolish “whiteness” and his objections to kosher toasters—Landry L. suggests that DSA and its allies should “actually instigate [a crisis] by deepening the contradictions in the current political order.” Even if Mamdani fails to deliver on his agenda, Landry L. argues, “his position as mayor will open possibilities that are bigger than a single mayoral administration in one city.”

These disputes reveal the extent to which Mamdani’s DSA coalition is fractured. His ideological allies cannot agree on the purpose of his mayoralty, or even whether it’s worthwhile. Is the mayor-elect merely an “organizer in chief,” or does he have the “radical potential” to spark a crisis that ushers in a socialist revolution? As the DSA takes power, New Yorkers should pay close attention to the answers.

Photo by CHARLY TRIBALLEAU/AFP via Getty Images