Ed Note: The word “fraud” is used liberally throughout the piece. Point of clarification—tho it should be clear from context, this word is usually a comment on the person or the work being substantively flawed or incompent, not necessarily a “cheater” in the AI usage or synthphobic sense.

Jamir Nazir takes full credit for lines of narrative prose like “The girl smiled like sunrise over a sink“ and “She had the kind of walking that made benches become men.“ Putatively qualified judges praised the story’s melodic voice and quiet authority. But rumblings started within weeks of his “The Serpent and the Grove” being announced as a regional winner and grand prize finalist in the Commonwealth Short Story Prize. Claims that he “cheated” by using generative AI. That his, and other works, should be disqualified, often on the say-so of AI detectors alone. I’m here to call bullshit, on all of the above. Without pity, without favoritism, and without face-saving.

The generative AI boom has led to all manner of startling developments and innovations, as well as anxieties around the issue of human obsolescence. Writers, and other creatives, have started splitting into at least two (if not many) camps over the question of AI using in the arts, in journalism, and in academia. I am not here today to (re)litigate those fights, at least not principally. I am here to call for a return to objective standards that would have seen pieces like Nazir’s get rejected before they even had the chance to be published by Granta (a writing publication which has since cut ties with Commonwealth over this scandal). Without AI detectors even being used.

Commonwealth, for their part, say they investigated. They claim to have determined that no AI was involved. Fine. Let’s say I grant that (even tho I’m dubious). All that means is a human wrote the utter dreck I quoted up top. And the process that was supposed to screen such dreck out…not only missed the problems, but celebrated them. At that point, the judges are objectively just as bad at their task as Nazir was at his. There’s no point debating whether he used a tool, or whether another tool could be reliably used to determine whether he did. I don’t want to outsource my judgment to AI detectors. You shouldn’t either. Particularly not when objective criteria exist.

Thesis

The coverage of the Commonwealth scandal doesn’t end with Nazir. Also implicated are Sharon Aruparayil, the putative writer of “Mehendi Nights”; and John Edward DeMicoli, of “The Bastion’s Shadow.” Other scandals include: Simbarashe Steyn Kundizeza, of “The Blue Fighter,” Thélyson Orélien of the book C’était ça ou Mourir (“It Was This or Death”), Mia Ballard of the novel Shy Girl, and Jerry Falade of the manuscript for the aborted novel Call Me, I’ll Hide the Body. A lesser known name I’ll add is Chanel Sutherland, the 2025 overall winner of the Commonwealth Short Story Prize for her story “Descend.” As people dug into Commonwealth’s back catalog, they found that even previous cycles had very suspicious winners. The press frames these as “cheating” scandals. I contend that these are actually quality-control failures.

For precision, I will admit that the quality claim is best limited to Aruparayil, Ballard, DeMicoli, Kundizeza, Nazir, Orélien, and Sutherland for now. All their works are public, Falade’s is not. I will judge each of the published works by the criteria I’ll outline below. Falade’s manuscript is unpublished, so he is technically more of a process case, but still worth unpacking. Other figures I’ll touch on will include Rie Kudan, Steven Rosenbaum, Olga Tokarczuk, and H.M. Wolfe. All but Wolfe have openly admitted some level of AI usage. All but Rosenbaum have been deemed qualified successes.

It is my contention that—beyond the wrongness of outsourcing our meritocratic judgments to AI detectors—we must be willing to state and defend standards of merit. Harold Bloom, whom I consider an important influence on my own professional critic work, was far better at the latter than the former. The Commonwealth judges, I would argue, suck at both. Various journalists, tech workers, and internet randos are further muddying the waters by leveraging AI detection scores in place of real aesthetic arguments and artistic or journalistic standards. The language of “they cheated” won’t work, especially not in cases where “cheating” isn’t even a valid option, like a decentralized self-publishing ecosystem or marketplace that never promised all of the content was “100% human” to begin with. We must be honest and do better.

Seven Tests, Derived

Harold Bloom told readers aesthetic value exists and that they must choose it over ideology. What he rarely did was show the distinction on the page. I’ll do that here. But not by inventing a rubric. Rather, I’m collecting what writers and critics from the time of Shakespeare to our present day (and the likes Stephen King or Cormac McCarthy) keep demanding. From rival premises to universals. The result? Each of the most consistent and durable demands gets turned into a test, a criterion. And they combine into a seven point criteria of objective merit. Independent traditions arrived at convergent truths separately. But there is still a fact of the matter.

Each piece is judged, and earns its place, if at least four of the criteria hold true within it. This explains why canonical works are judged as they are, including their known flaws. The criteria can be failed, so it isn’t a rubber stamp. And it can be applied the way David Hume’s “true judges” apply taste: with delicacy, practice, comparison, and freedom from prejudice. Hume held that taste is sentiment, yet that the “joint verdict” of such judges is the standard. He also thought work that survives long enough shows what that standard is. Bloom did too. As did Roger Scruton, and many others.

The measures are…

1. Specificity. Anton Chekhov told his brother in 1886 to seize small details so a reader closing their eyes gets a picture, and offered a bottle shard on a mill dam. King, writing a century later from a different tradition, says good description is a few well-chosen details standing for the rest. The test: could a sentence in the given work move to another work unchanged? If it can, it’s not specific (nor great).

2. Figures that cash out. Good metaphors and similes must earn their keep. Chekhov succeeds by comparing glass on a dam to a "bright little star"—a comparison grounded firmly in the scene's physical world that accurately reflects the moonlight. George Orwell warned against the opposite, saying stale, lazy metaphors that are just filler and fail to evoke any real image. The test: does the comparison actually illuminate its subject, and can the reader easily picture the image being used?

3. Function. Every part of a story must serve a purpose. Edgar Allan Poe demanded that every single word work toward the “pre-established design,” and King said rewriting means stripping away everything that isn’t the story. Aristotle made the same structural point—a true work is tightly built so that removing any piece disrupts the whole. The test: make one or more cuts. What breaks?

4. Vital repetition & exposition. If details and pacing changes actually enact something, serve a need, then they have function. “Tomorrow, and tomorrow, and tomorrow” drags time out. Other types of repetition or exposition that merely fill time aren’t necessary. King warns about the writer who grows enchanted with their own descriptive powers or ability to meander. The test: (how) can we say what’s being said differently? What is lost if we change it?

5. Earned ending (form). A story’s ending cannot be pasted on from the outside; it must be the inevitable pressure-release of everything that came before. As Aristotle noted, the resolution has to follow by logical necessity rather than authorial convenience. The test: is this conclusion the logical destination born from the characters’ early choices, or could you swap it out? And does changing a major early decision cause the whole story to collapse?

6. Cognitive footprint (yield). Great literature leaves an aftertaste of truth—it alters how we see the world long after we close the book. As Samuel Johnson and Harold Bloom observed, lasting art succeeds because it offers a true representation of human nature, marrying emotional impact with deep insight. The test: What do you know or feel afterward that you didn't before? And did the drama of the scenes actually earn that realization, or did the narrator just stand on a soapbox and announce it?

7. Durability. Johnson’s only test for works whose excellence is gradual and comparative is “length of duration and continuance of esteem.” Hume agrees, though he concedes that time is little use for new work. For new work, comparison against survivors substitutes. The test: does it survive rereading? And does it fare well beside its beloved peers in the same temporal epoch?

Let me address a few objections straight away. Firstly, why canonical violations don’t sink the list. Johnson called Shakespeare the poet of nature, then faulted him for quibbles, for carelessness with plot, and for a lack of proportion between words and things. A flaw in a great work is information about the flaw. My claim against these stories is about density: when the figurative and aphoristic layer fails several tests at once, repeatedly, that is a pattern.

Secondly, the list isn’t circular. We’re not triangulating to validate works that existed before AI, nor are we assuming one man’s personal canon is the basis of merit. The sources predate the current fight by centuries or even millennia. The tests can reject a work that’s been on bestseller lists or taught at famous universities. The tests also weren’t tuned to detect LLMese, and they will reliably fail human writers who fall short—which is the point of using them. Lastly, these tests govern execution, not genre or temperament, where Hume allows blameless variation.

I’ll be honest about what I don’t claim. Passing four doesn’t make a work great. Rather, failing a majority is strong evidence that a work isn’t. This is like how some entry level exams exist more to separate the wheat from the chaff. Not all wheat will produce equally good or great bread. Yet I will argue that the works we tend to value…tend to pass more than four and/or have great reasons for taking the liberties they do. I would dare say again that most of the scandals come down to bad works being rewarded, not AI being used.

Sample Application: Nazir, “Serpent”

Passes: plot mechanics (loose nail, lifted plank, cleared scrub, the lie about seeping water) and the dialogue, just barely. But it fails multiple elsewhere. Test 2: the sink isn’t in a story world of carried water, no fan and no bulb. Nazir’s explanation, car polish on his mother’s sink, never reaches the reader. The benches line decodes only if the reader supplies a premise the text doesn’t. Test 4: at least eight strict “not X but Y” constructions, plus comparative variants. Tests 5 and 6: a long coda that states the theme, aphorisms that interrupt scenes, and Zoongie as a motive object.

Let’s dig a little deeper. “She had the kind of walking that made benches become men.” Zoongie introduction. This decodes only if the reader supplies a premise the text never states. It’s just gibberish, and none of the judges flagged it. Meanwhile, journalists and casual observers rightly point out it’s slop, but use that to levy the “cheated” frame rather than address aesthetic merit and universal standards. We can continue on with more excerpts.

“Sun on galvanise is a cruel instrument.” Opening of the shack scene. Tests 2 and 4: instrument, drumskin and “porridge skin” all arrive within three sentences. “Not the bees’ neat industry […], but a belly sound.” First paragraph. Test 4: the first of at least eight not-X-but-Y constructions. “Doing is a treacherous bridge.” After the well rescue. Test 3: an aphorism that interrupts the scene to generalize. “Laughter can cut a hush, not cure it.” Puttie in the yard. Test 3: same. “The serpent in the grove was never only a snake.” Coda. Test 6: announces the theme the scenes already dramatized. “A story is a well. It eats sound until somebody throws a rope.” Coda. Tests 5 and 6: closing metafiction after the climax. It’s didactic and hollow.

My contention has always been that this story was objectively bad. That it should have never won its regional prize, let alone the Grand Prize. I didn’t need an AI detector to tell me this. I needed, to quote Sherlock Holmes in the 22nd Century: “eyes and brains, dear Watson. Eyes and brains.” I guess the Commonwealth judges forgot theirs.

Ballard and Shy Girl

Mia Ballard is a mostly self-published author who rose to prominence via Amazon’s KDP algorithm and social media. In the latter case, she was particularly viral on the “BookTok” wing of the platform TikTok. Her latest novel, Shy Girl, was originally self-published in February of 2025. After successful sales and internet virality, Hachette bought US rights in July 2025 and republished under their UK arm in November of 2025. As part of these processes, authors routinely have to sign contracts affirming that the manuscript is all their own work, and they are solely responsible for it.

YouTuber Frankie’s Shelf posted a nearly three hour video review and deep dive regarding Shy Girl in January of 2026, entitled “i’m pretty sure this book is ai slop.“ The YouTuber recounts endless repetition of “sharp,” heavy rule-of-three constructions, and other issues. A few months later, the New York Times investigates, and cites a Pangram readout of “78%” as a sign that most of the manuscript was AI generated. Hachette cancelled the US publication, and discontinued UK sales, in March of 2026. Hachette says it requires submissions to be original and AI use disclosed. Ballard admits, both in a since-deleted YouTube comment and in correspondence to NYT, that AI was in the manuscript. However, Ballard claims an “editor” she hired must have inserted AI into her work, not her.

Look, the standards and criteria I detailed are applicable to all written fiction. Frankie’s Shelf applied their version of them in a nearly three-hour deep dive into Ballard’s Shy Girl. That book would have been a strucural failure even if Ballard hadn’t admitted the draft she submitted was AI. Jezebel was one of the few publications, besides my own, to rightly put together that Ballard submitted content she’d warranted was wholly her own when, by her own later admission, generative AI was in the manuscript.

Whether her “editor” story is the bullshit it sounds like or not, Ballard has basically committed to, at minimum, a contract breach and an acknowledgement that the text was AI. The idea that she’s not responsible because some phantom “editor” did it is as laughable as a student not being responsible for their failed assignment because their dog supposedly ate it the night before. These excuses, like these works, are total amateur hour. It blows my mind that anyone views this as some injustice being done to Ballard, rather than a failure of her own making. Poor quality writing isn’t a crime, but lying counts as a basis for contract fraud and nullification.

Thélyson Orélien — Another Arday

Haitian-born Quebec novelist Thélyson Orélien is the credited author on multiple books and articles. His career stretches back to 2006, tho most of his success was recent, roughly in the mid 2020s. He received some token—a word a use deliberatively and advisedly—awards in Canada, France, and Switzerland prior to his current fame, mostly for his work either as a translator or as an “emerging immigrant writer” working in a non-native language. His credited book C'était ça ou Mourir (“It Was That or Die”) rocked him to success after he began to get positive reviews and awards chatter in the French-language and Quebec presses.

After the novel became a massive literary sensation in France and Canada, selling tens of thousands of copies and winning prominent accolades like the Prix du roman Fnac, accusations erupt in late September 2026. An anonymous group on X (former Twitter) accused Orélien of using generative AI to write the book. Journalists ran text excerpts through AI detection software such as Pangram, pointed to repetitive structural tics and stylistic patterns, and then started digging into his back catalog.

In short time, it was confirmed that Orélien had plagiarized prominent sections of his personal blog, his previous stories, and his paid journalism. This, in concert with the AI allegations, led France’s prestigious Goncourt Prize to immediately dropped Orélien from its running list. Investigations revealed that Orélien had a long history of past plagiarism, having lifted entire sentences and columns from older essayists, international blogs, and 20th-century books (such as those by Louis Pauwels and Jacques Bergier) over a span of years.

Orélien vehemently denied that AI wrote his novel, claiming the accusations were slanderous and racially motivated. He argued that the rhythm and imagery came naturally from his Haitian and Caribbean heritage, though he later admitted to using AI tools “here and there” strictly for research. He also admitted to past instances of plagiarism. Following internal reviews, metadata analysis showing suspicious digital markers, and the compounding plagiarism scandals, his Canadian publisher, Éditions du Boréal, officially cut ties with him.

My general sense is that Orélien is objectively a fraud in the same vein as a Jason Arday (the late Cambridge academic who was outed as a liar and plagiarist). I find his claims of being racially targeted both specious and offensive, just as I did in the Arday case. I take no great pleasure in stating that these men, who happen to share my skin-tone, were frauds. But anyone who asserts that their wrongs ought to be excused based on skin-tone is making a prima facie racist and insulting argument. Plenty of black writers have managed to create incredible works that passed the same standards as any other person. Alexandre Dumas, a black Frenchman, was the author of The Three Musketeers and The Count of Monte Cristo (among other hits). Dumas is still widely considered one of the Top 10 greatest French-language writers ever.

Disclosed-Use Foils

The comparatively minor literary stories surrounding Rie Kudan and Olga Tokarczuk offer a vital contrast to cases of outright deception or shortcut-seeking fraud. Rather than trying to pass off a machine-generated product as wholly human-crafted, both authors represent a transparent, critically engaged model of “disclosed use.” Neither cheaters nor bad writers, they demonstrate how established artists can interact with modern computational tools without sacrificing their artistic integrity.

Fraud relies on concealment—pretending a text is something it is not. Both Kudan and Tokarczuk openly and voluntarily discussed their processes. Kudan explicitly stated in her acceptance press conference that about 5% of her Akutagawa Prize-winning novel was lifted verbatim from ChatGPT. Tokarczuk openly addressed her research workflows and brainstorming methods in public interviews.

Neither author outsourced the core creative soul of their work. Tokarczuk firmly clarified that she writes completely alone and utilizes large language models strictly as a digital research assistant and fact-checker—a modern evolutionary step beyond library archives or Google searches. Kudan used ChatGPT not to write her story for her, but as an interactive sounding board to mimic and probe the psychological alienation of a technologically saturated world.

Why Disclosure Is Essential

• Preserving Trust and Consent: Disclosure forms an ethical contract between the writer and the reader. When authors are upfront about where and how technology intersects with their work, readers can evaluate the art honestly.

• Preventing the “Witch Hunt” Culture: Secretive or deceptive practices fuel toxic online panics and weaponized AI-detector “witch hunts” against innocent writers. Openness establishes a healthy precedent, separating deliberate fraud from legitimate artistic experimentation.

• Allowing Process Debates: If a writer opts to use AI and their audience consents, the question stops being one of “did you use X tool” but simply “is this work good?” Sure, there’s always the possibility that open AI usage will be judged differently than concealed usage or zero usage, but this tradeoff is worth it. We should be able to assess whether someone’s work benefits from the tool or not.

Comparison to Other Controversial Cases

When compared to other high-profile figures caught in AI scandals, Kudan and Tokarczuk’s work highlights a stark difference in intent, control, and literary merit. Let’s look at some key examples.

In the press coverage of his book The Future of Truth, Steven Rosenbaum suffered a catastrophic embarrassment because he blindly trusted AI to generate factual quotes and citations, resulting in fabricated data in a book about truth. By contrast, Tokarczuk uses AI strictly as a preliminary search engine while strictly verifying her information independently, avoiding the trap of unvetted machine falsehoods.

Writers who panic-hide total AI-generated prose usually produce homogenized, cliché-ridden filler, as Ballard is accused of. On the other hand, Kudan and Tokarczuk are master stylists of international acclaim. Kudan’s integration of AI was deliberately chosen to reflect the novel’s thematic anxieties about a dystopian, AI-driven society. The AI text within her book served a specific, controlled, and ironic artistic function rather than masking a writer’s inability to write.

The controversy surrounding Daggermouth—a viral, best-selling dark dystopian romance novel on BookTok by indie author H.M. Wolfe—exploded when researchers from Stony Brook University used the AI-detection tool Pangram to analyze a massive dataset of Kindle ebooks. The software scored Daggermouth at an astonishing “60% AI-generated,” prompting a high-profile feature in The Atlantic that thrust the romance publishing world into chaos. Critics and online sleuths pointed to narrative inconsistencies—such as shifting character ranks or unaddressed injuries—as “proof” of machine authorship.

However, the publisher and author forcefully rejected the claims, with Wolfe providing extensive drafts, writing logs, and notes to prove human creation. Prominent literary figures and industry analysts, such as agent Nathan Bransford, sharply condemned the reliance on algorithmic detectors, criticizing outlets like The Atlantic for treating unverified detection software as definitive journalism. The Daggermouth clash laid bare a terrifying industry reality: in the absence of genuine proof, the weaponization of flawed AI detectors can easily trigger reckless witch hunts against human authors.

It is telling that harsh critics of AI usage briefly tried, and failed, to ruin Tokarczuk’s reputation by misrepresenting translated remarks she made about her process in the press. Kudan, for her part, has been most unscathed by the press. She’s clearly buoyed well by the inherent quality that’s evident in her work, and even the most rabid synthphobe would struggle to justify calling her stuff “slob” as any kind of objective meritocracy assessment. She and Tokarczuk prove that the threat of AI in literature isn’t the technology itself, but the intent behind it. By keeping the human artist firmly in the driver’s seat and practicing radical honesty, they show that tools can be integrated thoughtfully without compromising true literary brilliance.

Gatekeeping Failure

The Commonwealth Foundation runs on an honor system, with writers’ statements and personal notes as the check. They do not believe in using AI detection. They have called their process “robust” and “fair.” I will only give the credit on the detector piece. Commonwealth’s Short Story Prize, like the Caine Prize, is for “emerging writers” from across the world, with a focus on current or former UK territories (in Commonwealth’s case) and especially African diaspora writers (in both, particular Caine’s).

Ellah Wakatama, the chair of the Caine Prize, expressed shock and dismay that The Atlantic chose to publish a hit-piece and taint the prize winner, Kundizeza, before giving the prize's official investigative process time to finish. She criticized the rush to judgment, noting a lack of independent third-party research or "real human investigation" backing up the software's sweeping algorithmic claims. I can again give partial credit to some of this. I am famously no fan of AI detectors, including Pangram. I tend think there’s a pernicious, even unethical, way that Pangram seeks to launder its accusations—and its own reputation—through stories like Kundizeza’s scandal.

That said, I am firmly of the opinion that Kundizeza—like Aruparayil, Ballard, DeMicoli, Nazir, Orélien, and Sutherland—is probably just a bad writer. Concerns have swirled for years, even prior to any AI scandals, about how the Caine Prize selects for maudlin stories full of regional stereotypes and hackneyed tropes. Commonwealth likewise seems to prioritize the optics of awarding people from the global south with money and prestige, totally divorced from objective standards of merit. The right frame isn’t a “cheating” scandal a la student cunning a university, but something closer to Jason Arday at Cambridge. Most people with working eyes and brains agree that Arday was a liar, a fraud, and a terrible academic. No AI required, on anyone’s end.

Jerry Falade’s manuscript, which we obviously don’t have access to, was flagged before the deal closed. I hesitate to call this a “failure” of gatekeeping when the gates ostensibly held. No fraudulent or feckless text made it to publication. Still, a 7 figure bidding war had apparently popped off prior to the text getting flagged, which is part of why the piece became a story in the first place. Test readers, for whatever it’s worth, did say it had hallmarks of LLMese, and simply wasn’t very good. Observers can draw their own conclusions about what it means that millions of dollars were offered up for a manuscript most of the people involved clearly hadn’t read.

Rival Frames

Let’s address some room-elephants. First up, detector reliability: Vauhini Vara, the Atlantic writer, cites a false-positive rate of about 1 in 24,000 and says she could be wrong. The Commonwealth Foundation, the Caine Prize chair, and defenders of specific authors reject detector-only evidence. I sympathize with this, but I will grant the very basic point that most AI detectors are accurate enough to tell the difference between “100% AI” and “100% human” (i.e. no AI). Results of detection software shouldn’t be the controlling factor, but I think anyone who hinges their reputation on “I was falsely accused” is already fighting a losing battle.

No, Pangram (the purported “best in class” AI detector) isn’t as accurate as claimed. Multiple independent analyses on Substack and elsewhere have called its own claims about its products into question. I treat “a false-positive rate of about 1 in 24,000“ as vacuous puffery akin to Papa Johns’ “Better Ingredient, Better Pizza.” If you blindly trust it over your own judgment, I have a water-well to sell you in the Arabian desert. But again, it can do what the synthphobes basically want it to, which is separate content that’s heavily or wholly AI from content that’s non-AI (or close enough to be marginal). However, I don’t think AI usage itself is some “pass-fail” binary. Kudan and Tokarczuk prove as much.

Second, I’ll bite some bullets on race and region. Most—tho certainly not all—of the accused are black and/or global south writers. Some don’t speak, or write, predominantly in English as their first language. Many were “emerging” writers without much in the way of mainstream visibility prior to their scandals. But all are, at least ostensibly, of average or above average intelligence. All wrote with the possibility (maybe even likelihood) of English-market publication and success in mind. They knew, or should have known, that their work would be judged by more than the gatekeepers they submitted to. They went through with the dubious works and schemes anyway.

The notion that AI detectors are racist seems dumb to me. It’s a just-so story that, similar to the Arday-Orélien claims of racially motivation harassment, seems to license a bigotry of low expectations. And besides, detectors aren’t required, as I keep saying. Each and every one of these pieces failed on the page. If someone wants to claim that black people, or non-Western people, cannot meet universal standards, I will again call them a low expectation bigot. I will then cite Dumas, and Maya Angelou, and James Baldwin, and Frederick Douglass, and Ralph Ellison, and Coleman Hughes, and Langston Hughes (no relation)…and maybe, just maybe, even myself. And I will dare that bigot to keep saying people who look like me can’t pass the same tests.

Conclusion

The language of “they cheated” is used to treat what is actually a conflict over merit into a nebulous question of rules-following. But as we can see from these cases, the “rules” are sometimes structurally unenforceable or non-existent. If writers disclose AI usage and don’t try to leverage gray areas into lies of omission, they and their work can be judged on the merits. Likewise, even for those who may well have “cheated” with AI, we can and should still use meritocratic standards of aesthetic judgment to determine whether their works are good or bad. No detector required.

What we cannot, and must not, do is outsource our judgments. Not even, sadly, to the gatekeepers who are supposed to do quality control before works are presented to us as notable or “worthy.” Sometimes judges get it wrong. Sometimes publishers are willing to throw millions at a manuscript they hadn’t bothered to read. When these things happen, we must rely on the same objective metrics that men from Aristotle to Harold Bloom have intuited (and iterated on) to ascertain quality.

I don’t claim to be either of those men, let alone infallible, but I have eyes and a brain. I know how to judge art. Criticism has literally been a job of mine, paid per piece, for years now. I don’t need a machine to tell me what’s good or bad, and neither do you. All we need is standards, scrutiny, and the scruples to say, “I don’t care how this was made, it sucks actually.” (Or, conversely, “it rules actually.”) All based on real criteria, not vibes or tokenism. Because people who look like me ARE capable of greatness, and I refuse to pretend we’re not just to serve some pernicious political agenda. Hopefully you feel likewise.

Secondly, one time donations are also availaible via Cash App ($Edokwin), PayPal, and/or Venmo (@Edokwin).

Thirdly, anonymous donations and alternative subscriptions are available via at Buy Me a Coffee!

Last but not least, you can schedule a meeting with me via Setmore. If you need my professional advice, or want to work with me on a project, that’s your best bet.

Thanks as always. Hit me up in the comments with smart replies!

See ya next time.

Sources

• Authors Guild. “Shy Girl and the AI Authorship Question.” Authors Guild, 2026, authorsguild.org/news/shy-girl-and-the-ai-authorship-question.

• Commonwealth Foundation. “The Serpent in the Grove.” Commonwealth Foundation, 2026, commonwealthfoundation.com/the-serpent-in-the-grove/.

• Eastern Herald. “Caine Prize ‘Blue Fighter’ AI Allegations Review.” Eastern Herald, 12 Oct. 2026, easternherald.com/2026/10/12/caine-prize-blue-fighter-ai-allegations-review/.

• Jezebel. “AI Book Publishing: Shy Girl and the Hachette Controversy.” Jezebel, 2026, www.jezebel.com/ai-book-publishing-shy-girl-hachette-withdrawn-canceled-controversy-slop-writing-mia-ballard-passages.

• KERA. “SMU Student Loses Book Deal Worth Over $2 Million After Allegations of AI Use.” KERA News, 13 Aug. 2026, www.kera.org/2026/08/13/smu-student-loses-book-deal-worth-over-2-million-after-allegations-of-ai-use/.

• Anderson, Sonja. “ChatGPT Helped Write This Award-Winning Japanese Novel.” Smithsonian Magazine, 24 Jan. 2024, www.smithsonianmag.com/smart-news/this-award-winning-japanese-novel-was-written-partly-by-chatgpt-180983641/.

• Pearson, Jordan. “Winner of Japan’s Top Literary Prize Admits She Used ChatGPT.” Motherboard / Vice, 19 Jan. 2024, www.vice.com/en/article/rie-kudan-akutagawa-prize-used-chatgpt/.

Welcome back to Rogue Reviews, the segment where I do the kind of criticism so risky that mainstream presses won’t publish it. Today’s topic is Lanterns, the HBO (Max) series by Damon Lindelof, Chris Mundy, and Tom King. A DC Studios superhero mystery show led by Aaron Pierre as John Stewart, with Kyle Chandler as his mentor/rival Hal J…

Read more

This piece takes the term further than Stanovich did. My claim is that dysrationalia is not a rare diagnosis but a widely distributed deficit, that part of it is learned socially at scale, and that modern information environments amplify both its spread and its cost. Some of my argument is cautious. The evidence for a general decline in rationality—a la the movie Idiocracy—is debatable (maybe even thin); whereas the evidence that environments change who suffers from the deficit is compelling.

Some of what follows here will be speculative, and flagged as such. A short hypothesis about autism and logical consistency comes early, because it suggests the deficit has a mechanism and not only a symptom list. I will almost try to keep the jargon and the syntactic complexity under control. Like the old saying goes, if you can’t explain it simply, you don’t understand it well enough, right?

The Backstory

Stanovich coined “dysrationalia” in a 1993 Journal of Learning Disabilities paper as a new discrepancy-based category: the inability to think and behave rationally despite adequate intelligence. He framed it as a “demonstration proof” meant to force more careful thought about how IQ figures into learning-disability definitions, not as a diagnostic curiosity. His concern was correcting people who assumed IQ and rationality are not only correlated, but that they never separate; that the “smarter” we are in terms of raw IQ or G factor, the less susceptible are to irrational thought.

His later work (What Intelligence Tests Miss, 2009; The Rationality Quotient, 2016) showed rational thinking is surprisingly independent of intelligence, with only small-to-medium correlations. I think his findings hold up rather well. We can see ample evidence that high IQ actually makes someone better at hiding, or post hoc rationalizing away, their own irrationality, even from themself.

My extension is from individual differences to population-level amplification, environmental cost, and trainability. I present it as an extension, not a novel invention or grand discovery. Stanovich never proved a population-level claim, but I do think his findings imply one. And credit where it’s due elsewhere in the field: Kenneth Kavale’s objection and Robert J. Sternberg’s call for a fuller conceptualization are part of the term’s history. I try to incorporate and synthesize the best ideas from all three.

Rationality Restated, Deficient Defined

Here is my working definition of “rationality”: the ability to think clearly, comprehend old and new information, and to avoid logical incoherence and contradiction. I present this as the testable core of my work here today.

Stanovich offered a two-part standard which I’ll be working with as well: epistemic rationality (beliefs which track the world) and instrumental rationality (actions which achieve goals efficiently). Coherence is necessary for both but not sufficient alone. Beliefs can be coherent and false. Actions can be inefficient without us knowing it. And the epistemics can conflict with the instrumental, requiring a tie-breaker (which could itself be arbitrary or irrational). We must be mindful of this as we proceed.

Another important concept is the Tripartite Mind. We have three mind “types”: the autonomous mind (fast, automatic); the algorithmic mind (what IQ tests mostly measure); and the reflective mind (thinking dispositions, the decision to override a default response). Dysrationalia is found as a deficit in reflective-mind dispositions (fluid rationality) and/or in mindware. Consider crystallized rationality: what if there are gaps or arbitrariness in things like rules, strategies, known knowledge? Or contaminations of processing in mindware, ones that actively mislead us? All this despite adequate algorithmic capacity (IQ). We need to account for the deficit.

Let’s take a closer look at override failure, the reflective mind deficit. Your fast, automatic mind jumps to a conclusion, which happens all the time. Humans are designed to do this, and usually it’s adaptive. But what if the conclusion is wrong? Under epistemic rationality, your reflective mind is supposed to step in and stop your automatic mind—cut off the faulty conclusion. If it doesn’t, this is irrationality.

Let’s consider a shopping scenario. Someone bought a bat and ball, which cost $1.10 in total. We’re told the bat costs “$1.00 more“ than the ball. If we had to quickly guess how much the ball cost, many people’s fast mind would instantly guess “10 cents.”

But that guess would be wrong. If the ball were 10¢, then “$1.00 more“ would make the bat cost $1.10, thus making the total $1.20. The correct answer is that the ball costs 5 cents, since “$1.00 more“ would make the bat cost $1.05, and the total would equal the original $1.10 that we were told. This is not intuitive or fast for most people, but it is thoroughly rational. Thus we need error-catching and remediation.

Now, the error can be corrected in a number of ways, including talking it out or looking at the actual receipt. But a person with an override failure accepts the gut answer without slowing down to check the math, even if they have a high IQ. Worse yet…whereas this analogy is relatively simple and low stakes, we encounter dozens or hundreds of cases where people default to a fast, simple, or intuitive answer to a complex question and refuse to budge off that answer when confronted with counter-evidence. Consider people who talk past each other over objective facts in science.

Next, we’ll look at the mindware gap (missing tools). This is where we have the raw brain power to figure something out, but we simply never learned the specific logical rule or probabilistic tool needed to properly evaluate it. Let’s try an example in understanding risk. A patient with potential illness might hear that a diagnostic test is 90% accurate and assume a positive result means that they almost certainly have a disease—without realizing they need to account for how rare the disease is in the general population. The patient isn’t dumb, not necessarily; they just lack the specific “mindware” tool. This is why it’s important for doctors to explain things in context.

Finally, we have contaminated mindware (bad tools, or viruses infecting them): learning rules, beliefs, or strategies that are actively harmful or illogical, leading you to bad conclusions. Consider the Gambler’s Fallacy or Pseudoscience Quackery. Someone adopts a belief that “a roulette wheel is ‘due’ for red because it landed on black thrice in a row.” This is a contamination, a learned “rule” that is mathematically false and leads to bad decisions. Compare with someone who learns that doctors or scientists sometimes withhold information or feel induced to recommend things that aren’t proven. This can create a contaminated “rule” where medicine or science is reflexively mistrusted by the public, with worse alternatives subbed in.

My point then is noting that mindware is learned, so part of the deficit is fixable. This supports the notion of trainability, and the environmental claims. Further, I argue rationality isn’t one thing, and that dysrationalia names a profile. The claim on offer then is that the profile dissociates from IQ.

CART (Comprehensive Assessment of Rational Thinking)

Since standard IQ tests do not measure rationality, Stanovich and his team created the CART to measure rationality directly. Whilst an IQ test measures raw processing speed, working memory, pattern recognition, and abstract reasoning (the algorithmic mind), the CART measures how well you actually process evidence, avoid bias, evaluate arguments, and make decisions in the real world (the reflective mind + mindware). By decoupling these and measuring them separately, we can understand the ways they differ, and how to best solve for deficits between them.

The resistance to cognitive biases can be understood as avoiding things like confirmation bias (where you only look for evidence that agrees with you) or framing effects (being manipulated by how something is worded). My current forthcoming work on the phenomena of “I’m right” and “myside” biases—especially in politics—applies to the former, as does my addressing “self-flattering narratives” for the latter.

Instrumental rationality is key for processing scientific & probabilistic reasoning. We can’t understand things like odds, risk, and basic scientific principles if we’re too irrational to recognize when we’re missing something, despite normal-to-high IQ. Likewise avoiding mindware contaminants. We must reject conspiracy theories, anti-scientific beliefs, or superstitious thinking—for reasons of epistemic rationality, not just social cohesion or conformity.

Because rationality consists of separate skills, you can be very good at one part of CART (e.g. probability math) while failing another (being vulnerable to confirmation bias). To iterate: epistemic rationality asks whether your beliefs are accurate to the real world? (Do you believe things because they are true?) And instrumental rationality asks whether your actions actually help you reach your goals? (Did you act efficiently to get what you want?) Furthermore, because rationality relies heavily on “mindware” (learned tools and strategies) and thinking habits (learning to pause and question yourself), it can be taught. Just like reading and arithmetic. Dysrationalia is fixable (treatable), it isn’t a permanent neurological defect.

The “Disability” Dilemma

Calling dysrationalia a “learning disability” is tricky. If it’s just something people were never taught, or a “normal” trait, we might be pathologizing it unfairly or prematurely. However, because irrationality causes massive real-world harm, treating dysrationalia as a critical skill deficit matters. Whilst Stanovich focused on individual differences between people, I argue that when society is full of highly intelligent people with low rationality, the environmental cost to society is huge: weak informational hygiene, poor public policy, financial crashes, and sociopolitical polarizations.

Here is one data point for the disability hypothesis: if dysrationalia is partly a failure to resist affect-driven inconsistency, then traits that dampen that pull should be associated with lower dysrationalia on at least some measures. Autism is a candidate. In a 2008 study, researchers found that autistic adults were less susceptible to the framing effect, choosing more consistently whatever the wording, and that they showed no differential skin-conductance response to gain and loss frames. The authors read this as a failure to incorporate emotional context into decisions; enhanced consistency, possibly at the cost of behavioral flexibility. These framing resistance results replicated in later studies, and one found that adolescents with autism gathered more information before deciding on a jumping-to-conclusions task.

My hypothesis here is narrow, at least for the moment. Some autistic cognitive styles resist one source of incoherence, affect-based framing. It is not that autistic people are generally more rational or “better” in a high-key sense. Framing is one bias among many; myside bias appears independent of intelligence and may behave differently. Still, the finding suggests dysrationalia may have a mechanism: that emotional context overrides stated premises. Reading this as a failure, rather than a strength, in allistic (non-autistic) thinkers could prove useful to understanding irrationality and dysrationalia as formal, diagnosable problems rather than “normal” traits.

Learning Disability Analogy, pros & cons

Learning disabilities are defined as selective deficits relative to measured intelligence; dysrationalia has the same shape. Dyslexia, for example, centers on phonological processing, not perception; it is highly heritable (estimates around 70%), and outcomes are moderated by home literacy environment and instruction. Researchers now use a multiple-deficit framework with gene-environment interplay.

A heritable or dispositional propensity plus an environment that sets how much it costs would fit dysrationalia like a glove. However, learning disabilities are defined by academic impact, and dysrationalia has none by definition. Kavale’s objection is that dysrationalia is closer to a thinking disorder, not a learning disorder. The dyslexia field has also moved away from IQ-discrepancy definitions, a move Stanovich himself championed, while Kavale and Forness defend discrepancy as a legitimate concept.

For these and other reasons, I use “learning disability” more as an analogy, and I accept the categories of “reasoning deficit” or “thinking disorder.” I define the deficit by profile: below-expected performance on rational-thinking measures relative to algorithmic mind ability. I do want to avoid medicalizing everyday error, so I will say clearly that this is a continuum. Just as someone could be “bad at math” without having per se dyscalculia, someone could be “bad at rationality” with having per se dysrationalia. Mostly what I want to push back on is the ambient idea that there’s nothing to see here at all, that it’s just “normal” to be irrational.

The Amplification Thesis

My sense is that irrationality is a distribution. My claim concerns the lower tail and its costs. I freely admit this is a controversial claim. I will try to make it falsifiable, scope it carefully, and separate evidence from inference.

Let’s begin with environments. We see a widening lower tail and rising costs since roughly 2010, tho not a decline from a golden age. The last century had Lysenkoism and Nazi eugenics, and a 1950s peak in conspiracy letters, with no overall trend from 1890 to 2010. Four survey analyses found no increase in conspiracy beliefs. Some Gallup paranormal items rose from 1990 to 2005. Baselines of rationality were never high, and it’s not my contention that we’re at a historical low.

In The Bias That Divides Us, Stanovich notes that unlike most cognitive biases, myside bias does not correlate negatively with IQ. High algorithmic intelligence does not protect against evaluating evidence in favor of prior beliefs; instead, it equips individuals with more sophisticated tools to post-hoc rationalize away contradictions. Whereas classical Festingerian psychology theory posits that “cognitive dissonance” (holding conflicting beliefs) produces mental discomfort that compels individuals to resolve the conflict (often by seeking truth or adjusting beliefs), we see that modern algorithmic and echo-chamber incentive structures alter this dynamic.

When social standing, group membership, or monetization depends on signaling team loyalty, resolving dissonance toward objective truth carries an actual social cost (ostracism, loss of audience). Digital platforms reward immediate validation and outrage, which minimizes the cognitive effort needed to resolve dissonance, effectively rewarding “dissonance tolerance” (living comfortably with logical contradictions as long as they serve group identity). It is worth pointing out that whilst psychologists understand myside bias mechanisms well, there is no cross-cohort time-series measuring "dissonance tolerance" or identity-protective post hoc rationalization (“copium,” as Zoomers and younger millennials might put it).

The OECD’s Program for the International Assessment of Adult Competencies (PIAAC) tracks real-world information-processing capabilities (literacy, numeracy, digital problem-solving). Critically, the proportion of adults in the top skill tiers remained relatively flat, while the proportion in the lowest proficiency bands grew, demonstrating a widening lower tail. Norwegian Registry (Røgeberg & Bratsberg, 2018): Studies examining the reversal of the Flynn Effect in Scandinavian cohorts show that within-family drops in cognitive performance point to environmental, not genetic, drivers (e.g., changes in education systems, media consumption, or reduced engagement with complex text/math).

Observing & Isolating Dysrationalia

Now I need to clarify some things, like dysrationalia isn’t. It’s not stupidity, nor ignorance, nor (necessarily) bad faith. Rationality and intelligence correlate only small-to-medium; a person with dysrationalia is nearly as likely to have above-average intelligence as below-average. Dysrationalia isn’t merely transitory irrationality. It is a prolonged and normalized/standardized tendency, and treating it is likely complex.

Ignorance is corrected by information; mindware gaps sit between ignorance and deficit, since the missing piece is a rule or strategy. Irrationality serves as a better explanation than malice (bad faith) for most of the problems we see, since many people show the pattern without malicious intent or behavior, including at higher IQ. Defining dysrationalia behaviorally helps us, since intent drops out of the analysis and no mind-reading is required. It also makes the diagnostic criteria more objective.

Observation standard: Firstly, my position is that motivated reasoning exists as one expression of dysrationalia. The test would be looking to see if dysrationalia appears reliably on content-neutral tasks like Cognitive Reflection Tests (CRTs), where the intuitive answer is wrong and reflection corrects it and on neutral syllogisms. The results should predict outcomes beyond identity. Secondly, we should restrict examples to formal standards such as logical validity, probability, internal contradiction of stated premises, failure to update when a premise is falsified, acceptance of non-sequiturs, inability to track a multi-step argument, comfort with conclusions that contradict stated premises, failure to update. Thirdly, we should pair each with a task from the CART or a CRT-style item, not just a G-loaded one.

Now let me state what follows if the thesis holds, aimed at institutions and information design. One descriptive claim is that reasoning deficits are widely distributed, including among the highly educated, and are not measured by IQ. Public discourse is conducted by people with the same myside bias as everyone else.

Since mindware is learned, the remedy is education and environment redesigns. One small study found debaters’ actively open-minded thinking rose over a semester while a comparison group stayed flat. Institutions should design information environments, school curricula, and professional norms so that coherence and updating are cheap and incoherence is costly again.

Conclusion

Dysrationalia is an old finding with a new application. Stanovich showed that intelligence and rational thinking come apart; the open question is whether the world we have built widens the gap. The evidence counsels care. Adult skills measures have slipped in several countries, and the strongest data on cognitive decline points to environmental causes that nobody has yet pinned down. What can be said with confidence is narrower and more useful: the deficit is common, it is not an intelligence problem, part of it is learned, and whatever is learned can be unlearned. If that is right, the policy question is not who should be allowed to reason but what kind of environment makes good reasoning cheap and bad reasoning costly again.

The test of this piece is whether its predictions fail. If reflective-thinking measures hold steady across cohorts, the thesis shrinks to a claim about distribution and cost. But I’d humbly argue that smaller claim is still worth making. Dysrationalia seems like an ambient problem that we either misread as “low IQ” behavior or treat as a non-problem for prudential reasons. I’m calling for a change. Addressing it as a “learning disorder,” even metaphorically, might just get us to a better understanding.

Secondly, one time donations are also availaible via Cash App ($Edokwin), PayPal, and/or Venmo (@Edokwin).

Thirdly, anonymous donations and alternative subscriptions are available via at Buy Me a Coffee!

Last but not least, you can schedule a meeting with me via Setmore. If you need my professional advice, or want to work with me on a project, that’s your best bet.

Thanks as always. Hit me up in the comments with smart replies!

See ya next time.

Works Cited

Bratsberg, Bernt, and Ole Røgeberg. “Flynn Effect and Its Reversal Are Both Environmentally Caused.” Proceedings of the National Academy of Sciences, vol. 115, no. 26, 2018, pp. 6674–78, pubmed.ncbi.nlm.nih.gov/29891660/.

Brosnan, Mark, et al. “Adolescents with Autism Spectrum Disorder Show a Circumspect Reasoning Bias Rather than ‘Jumping-to-Conclusions.’” Journal of Autism and Developmental Disorders, vol. 44, no. 3, 2014, pp. 513–20, doi:10.1007/s10803-013-1897-5.

De Martino, Benedetto, et al. “Explaining Enhanced Logical Consistency during Decision Making in Autism.” The Journal of Neuroscience, vol. 28, no. 42, 2008, pp. 10746–50, doi:10.1523/JNEUROSCI.2895-08.2008.

Džafić, Alma, and Nikola Erceg. “Agree to Disagree: The Role of Competitive Debate in Fostering Adolescents’ Actively Open-Minded Thinking.” Judgment and Decision Making, 2026, doi:10.1017/jdm.2026.10048.

“The Enduring Allure of Conspiracies.” Knowable Magazine, 2021, knowablemagazine.org/article/mind/2021/the-enduring-allure-conspiracies. Accessed 22 Oct. 2026.

Jänsch, Claire, and Dougal Julian Hare. “An Investigation of the ‘Jumping to Conclusions’ Data-Gathering Bias and Paranoid Thoughts in Asperger Syndrome.” Journal of Autism and Developmental Disorders, vol. 44, no. 1, 2014, doi:10.1007/s10803-013-1855-2.

Kavale, Kenneth A., and Steven R. Forness. “Discrepancy Models and the Meaning of Learning Disability.” Technological Applications, edited by Thomas E. Scruggs and Margo A. Mastropieri, Emerald Group Publishing, 2001, pp. 187–235. Advances in Learning and Behavioral Disabilities 15, doi:10.1016/S0735-004X(01)80011-2.

National Center for Education Statistics. “Adults Score on Par with International Average in Literacy Skills, Below International Average in Numeracy and Problem-Solving Skills in Survey of Adult Skills.” Press release, Institute of Education Sciences, 10 Dec. 2024, ies.ed.gov/learn/press-release/u-s-adults-score-par-international-average-literacy-skills-below-international-average-numeracy-and. Accessed 26 Oct. 2026.

Shah, Punit, et al. “Emotional Decision-Making in Autism Spectrum Disorder: The Roles of Interoception and Alexithymia.” Molecular Autism, vol. 7, art. 43, 2016, doi:10.1186/s13229-016-0104-x.

Snowling, Margaret J., and Monica Melby-Lervåg. “Oral Language Deficits in Familial Dyslexia: A Meta-Analysis and Review.” Psychological Bulletin, 2016, doi:10.1037/bul0000037.

Snowling, Margaret J., et al. “Defining and Understanding Dyslexia: Past, Present and Future.” Oxford Review of Education, vol. 46, no. 4, 2020, pp. 501–13, doi:10.1080/03054985.2020.1765756.

Stanovich, Keith E.

• The Bias That Divides Us: The Science and Politics of Myside Thinking. MIT Press, 2021.

• “Dysrationalia: A New Specific Learning Disability.” Journal of Learning Disabilities, vol. 26, no. 8, 1993, pp. 501–15.

• What Intelligence Tests Miss: The Psychology of Rational Thought. Yale UP, 2009.

• The Rationality Quotient: Toward a Test of Rational Thinking. MIT Press, 2016.

Sternberg, Robert J. Educational Researcher, vol. 23, no. 4, May 1994, pp. 11–21.

“Three in Four Americans Believe in Paranormal.” Gallup, 16 June 2005, news.gallup.com/poll/16915/three-four-americans-believe-paranormal.aspx. Accessed 22 Oct. 2026.

Uscinski, Joseph, et al. “Have Beliefs in Conspiracy Theories Increased over Time?” PLOS ONE, vol. 17, no. 7, 2022, e0270429, doi:10.1371/journal.pone.0270429.

One of the ten biggest paradoxes about this place is that all of the above are true, even as some feel mutually exclus…

Read more

Months later, gray “Verified Creator” checkmarks started appearing beside famous names: , , , , . They’re documented nowhere. They were announced…nowhere either. And they cut against Substack’s own promise that “paying readers, not Substack,” decide who gets status and badges on their platform. That promise came back in November 2022, when the Bestseller badges first dropped. They where promoted as an alternative to capricious, black-box verification schemes at places like pre-Elon Twitter or the emerging pay-to-play rumblings at post-Elon X and even Meta. They were fair, or so the thinking went.

But then someone moves ably within that framework and gets flamed into oblivion. And months later, Substack decides, due to its ongoing fetish for celebrities, to quietly do a workaround. I don’t think the sequencing is accidental. My read? Substack watched what happened to Doyle and built a credibility badge that doesn’t advertise a celebrity’s paid-subscriber rank. As a corporate instinct, that’s almost defensible. Celebrity launches are sometimes a matter of damned if you do, damned if you don’t.

What isn’t defensible is the silence. On October 11 of 2026, I emailed Substack’s press inbox five questions and a deadline of October 16. I got nothing but a generic auto-reply on October 12th. The radio silence wasn’t just deafening, it matched a pattern I’d gotten from the vast majority, (tho not all) of the verified accounts that I reached out to. With Substack, I call it weaponized indifference. With the accounts, I call it a code of silence, one I suspect is encouraged by corporate. Cost of doing business.

Last year I wrote about the verified program, mostly just to explain that it exists and that seemingly no one was talking about it. Today I’m gonna tell you, through a mix of original research, extrapolation, and conjecture, how it actually works and why.

Receipts — Such That We Have Any

I’ll start by saying that I want to make the paper trail impossible to dispute before any argument. Whether I succeed in that is for you to judge. Let’s start with my original article and my calls, both public and private, for information on this program.

I talked about how and why verification was invented at social media firms. I noted, in my section “An Imposter Problem That Wasn’t“ that we don’t have some rampant issue of people impersonating celebrities here, at least not yet. Maybe that’s changed in the several months since the program started, but I doubt it. Substack is still a super niche platform, all thing considered. And it had other options.

In my “Ain’t Broke, Don’t Fix It“ section, I explained how the bestseller badges solved the problem that verification was developed to tackle on Twitter, Facebook, et al over a decade ago. All without Substack having to administrate it directly, since readers picked who the “real” version of someone was by paying for subscriptions. I noted how celebrities could promote their Substacks in un-fakeable external channels.

Imagine the alpha, the free publicity, and so on, from a Notable going on legacy media and mentioning their Substack page? They get subscribers (and money), sure, but the platform gets a whole new influx of visibility and audience. To quote moi:

Alas, that’s a choice and an ethos they ultimately seemed to increasingly renege on. As I detailed in the “Pattern of Opacity“ section, plus my platform accountability reporting before and since, Substack has been quietly disregarding their own rules and norms for at least a few years now. The introduction of Flower badges, undisclosed algorithm changes, the Native Sponsorships (ads) push, discovery tool removal, leaderboard changes, the Pangram integration, and so much more.

Questions, Questions — So Many Questions

I ended on the prophetically titled section “Questions Substack Won’t Answer,“ and the following…

Here are the questions Substack should answer:

1. What are the criteria for receiving a gray verification badge?

2. What role does Instagram, and its parent Meta, play here?

3. How many people have been granted these badges?

4. Why was this system created?

5. Why no public announcement?

6. Why isn’t it documented in help files or support pages?

7. How does this badge system align with the stated value that “paying readers, not Substack, decide who gets a badge”?

I also implored the public (i.e. my readers) to ask around and report back to me, either in the comments or in private. But being the journalist that I am, I obviously didn’t stop there. I asked dozens of verified accounts, plus the principals and/or their management teams, for comment either on the record or on background.

Most didn’t answer. I will repeat: most. Like, the vast, overwhelming majority. But I got a few nibbles despite the code of silence. It’s clear that Substack doesn’t want this talked about. It’s similarly clear that these accounts, tho “official,” aren’t being used directly by the named person or business per se. They’re managed, which is entirely common, even standard, for celebrities in the digital age. But it does have certain implications, which I’ll touch on more below.

Back to the questions. Let’s add in, and compare, the ones from my 2025 article with the email I sent Substack corporate:

“1. What are the criteria for receiving the gray Verified Creator badge?

2. What role do Instagram, LinkedIn, et al and their parent companies play in granting or validating it?

3. How many accounts currently hold the badge?

4. Why was it introduced, and why has it not been publicly announced or documented?

5. How does this badge align with Substack’s stated position that paying readers, not Substack, decide who gets a badge?“

Shorter, but largely the same basic ideas. I’ll tackle each so you have better context.

Criteria & Company Connection (#1 & #2, both lists)

Currently, all we know is that a verified badge can be attached, at Substack’s discretion, to any account and usually has a readout like the below:

I changed/added “Linkedin” because I’ve found examples of accounts linking back to a Linkedin (parent company Microsoft) rather than a Meta platform such as Instagram. I’ve even seen some linked to X (formerly Twitter), often politician accounts. Obviously, these tech companies—of which I’d naturally include Substack—all communicate with each other via back-channels. It would be nothing for Substack to get assistance, or just tips and tricks, on how to run a verification scheme from Meta, Microsoft, or even X Corp. Perhaps from former employees, perhaps from current ones. A lot of folks bounce around these firms throughout their careers, fun fact.

But that’s just explaining the how. The biggest question is “why.” Like, why not just run their program and stipulate that they verified the account in-house, with no external validator? My guess is twofold. First, it provides a bit of distance and plausible deniability. They can say they’re not running a verified program in the traditional sense, they merely ensured that this account linked an account that just happens to be verified on a separate platform (with a separate owner, separate staff, and its own in-house verifications). Second, it may look more “convincing” than simply saying, “we verified them,” and it promotes an account that the holder wants to promote.

How Many Verified Accounts Exist (#3, both lists)

This is actually the easiest question to answer with something non-specific and/or noncommittal, like “this program is ongoing and we onboard new accounts everyday.” I’m confident Substack could provide a numerical answer that would be roughly accurate, rounded to the nearest hundred. But even with that, admitting that they’ve done this for over a hundred people would be a significant confession in itself.

Without a number, Substack can always fallback on the notion that it’s a special exception they’re only doing for a small handful of people. The observable reality shows that’s bullshit, but they could play pretend. As it stands in October 2026, I believe Substack has applied the verified badge to at least 150 accounts, with plans to add another 100 to 150 within the next 6 months. I base on this on their longstanding pattern of reaching out to pre-existing Notables and getting them to join up.

Most of the major Substack creators who had a media or entertainment job prior to opening up shop here, all the way back to the late 2010s and early 2020s, were contacted by Substack directly. Matt Taibbi, Matt Yglesias, (lotta Matts,) Heather Cox Richardson, so on and so forth. Even in my neck of the woods, Atlanta journalists were asked to join and given upfront “advances”—sort of like author advances in publishing—on their future bestseller revenue, which they’d split with Substack for the first year.

All the verified program did is take the money out. Notables come, get verified, and get to do whatever they want. If they choose to monetize, fine. If they don’t, or simply fail to hit the 100 paid sub minimum for the basic bestseller badge, no sweat. But I’m maybe jumping ahead to #4, so apologies. Let’s cover that next.

Problems the Verified Program Solves (#4, both lists)

As noted above, the Doyle episode was probably the catalyst, amongst a number of other things. Her arrival, with a paid tier, 217K subscribers total (most of them free and email list imports), and a big red carpet rollout from Substack corporate, rubbed some people wrong. That led to the dogpile backlash that fueled her departure only a week later. Substack wants people like Doyle. But they don’t want controversies like hers. They don’t want Notables to feel unwelcome. So their solve is removing money.

Like I mentioned in previous essays, money sickness is endemic in arts spaces, where a creator charging at all can become a scandal. Critics framed Doyle’s debut as a “toxic” combination of celebrity and capitalism. Now let’s go back to those advances I mentioned. In 2021, Vox reported on a “Substack Pro” program and how it triggered a similar fight over money, audience share, strucural support, and who’s making it on a platform that ostensibly prides itself on an independent, community-led ethos.

McKenzie gave the Ankler an interview in June of 2025 dismissing outrage over celebrity arrivals as a “scarcity mindset.” I could say something about how leadership will talk about celebrities to friendly press and won’t answer a press inquiry about their own products, but let’s leave it aside. The point is, Substack wants celebrity users, and they’ll do all manner of things to get them. From upfront payments, to speed-running bestseller status, to these new verified badges. If celebs don’t want to charge for subscriptions, or fear blowback to doing so, the verified program is a cheap fix.

Why the Radio Silence? (#4, #5, & #6, first list)

This one’s kinda easy, given all the above. Substack is fundamentally unaccountable, and they certainly don’t like answering questions that will prove awkward or embarrassing. (Least of all to peons like yourself truly.) The verified program isn’t essential infrastructure, it doesn’t need to be outlined or delineated to the public, who won’t have access to it the way they do the bestseller badges or flower badges.

Should Substack be transparent? Sure. The same way oil companies shouldn’t pollute and industrial agriculture companies shouldn’t abuse land or animals. But reality is what it is. I can only report on what I find, not wave a magic wand to change it.

Verified Programs vs Values (last question, both lists)

I’ll be blunt as ever here. Let’s say the question is some variation of following: did Substack renege on their bestseller badge promise from 2022? The answer is obviously and indisputably “YES.” There’s no debating that point. All we have left to argue over is how we feel about that fact. I’ve said before that I’m not leaving anytime soon, certainly not over something like this. But that does raise another question: how many of us feel like that, about a lot of stuff, and what does it do to us? And to society, at scale? What happen when we all collectively shrug our shoulders at bad behavior? When we complain, in public or private, but do nothing otherwise?

Freddie deBoer, psycho tho he might be, warned Substack reps privately that some of the company’s recent moves would backfire. Even favored daughter Emily Sundberg has been iced out on occasion, when her question wasn’t conducive to the PR strategy that Substack wanted to push. I’m nowhere near the level of either person. I don’t have nearly the sunk costs. I could leave, and I might just yet, tho not over a badge.

Still, I know that lying is immoral, as is being unaccountable to your customers or the larger public. There are folks out in various corners of Substack-land (LitStack, PhilStack, etc) who think I’m an asshole for holding the line on certain things, like plagiarism, or facts, or basic human decency. It is funny to see what people will rationalize as “just how things are” vs what they’ll castigate as a mortal sin.

Damned If They Do, and Why

Let me really build the defense. Substack’s strategy, even tho I disagree with it, has a logic to it. They need Notables to grow, and they need to grow as a way of making money. I don’t think they need infinite growth, and I don’t think their 2010s strategy should be their 2020s strategy, as regards creator acquisitions. But let’s make the case anyway, just so we can say we were fair to the counterarguments, come what may.

Substack recruits top talent, coordinates celebrity launches, and helps fund the initial runs of certain creators (or at least used to). This means the people they bring have a reason to join, to stay, and to prosper. That means more money, more visibility, and more social cachet for Substack, at least in the long run and on average. (Some will still leave or drift regardless of initial incentives.) Having celebrities helps, and programs or processes geared to their proliferation are good for “everyone,” in a way.

I personally have, at least according to Substack, benefited from drafting energy, and subscribers, off the winds of bigger accounts like The Free Press and Jonathan Haidt. I can say definitively that accounts like Free Black Thought have helped me. Bigs help littles insofar as bigs are longstanding presences and create or grow communities that wouldn’t scale otherwise. They’re attractors, and anchors, like having a major brand name at a mall. Come for the Weiss sisters, stay for some guy named Quinn?

Problematic Program Payoffs

Now that we’ve setup something close enough to a steelman, let’s break it. First and foremost, the “Pro” program longitudinal results. Whilst the many creators who Substack paid came here, built out their brands, and stayed, we actually don’t know how many dipped after the checks stopped. I can tell you I know at least a few whose posting dropped off precipitously. Even with a bestseller badge, they weren’t motivated to keep up the output or the quality. Especially as Substack started reaching out to celebs who, unlike journos or authors, could bankroll themselves from the jump. “Pro” is largely a thing of the past now, and that’s a past-as-prologue sign.

Compare with verifieds. A common trajectory is that someone will join, post a few articles over a few weeks or months, then get bored and dip. Charli XCX—who, interestingly, chose to have her verified badge removed earlier this year—hasn’t posted since May of this year, and her post before that was in December 2025. Lizzo hasn’t dropped an article since March (tho she does post the occasional note, as recently as September). Other celebs, like Julia Louis-Dreyfus, turned on subscriptions and just earned a regular bestseller badge. Which is probably a net win overall.

Point being, most folks who are begged to come here will treat it like a fleeting trend and abandon it just as fast. Meanwhile, the program exists as an open secret and weird question mark, when a new entry in the Help Center page takes minutes to write and to explain. And hiding it to dodge backlash guarantees worse backlash later.

Most of all, I think the silence implies that Substack’s defense, such that they even have one formulated in their own minds, wouldn’t survive daylight. They certainly haven’t earned the benefit of the doubt, tho I did still try to see things from Substack’s perspective, and to outline the use cases and utility. Speaking of…

Politician Particulars

Another reason for the program, which I nearly neglected, is that some of the folks Substack has started reaching out to might legitimately and non-neurotically want to avoid charging money for content, on ethical or professional grounds related to their industry. I’m talking, ironically enough, about politicians. All jokes about their morality aside, folks like Josh Shapiro, Bernie Sanders, and so on probably don’t want the optics headache of explaining why they’re charging money for their thoughts online.

I could say it’s a somewhat arbitrary concern relative to hawking books or getting board seats from companies they used to regulate, but still. Verification can serve as a useful stopgap measure that ensures accounts are “real” without needing to be paywalled. I can understand the desire to find a compromise in this way.

Politicians such as Buttigieg and Shapiro also do, in some cases, have the odd habit of staying longer and being more active than random celebs. So for that reason alone they seem worth catering to, even if doing so runs counter to previous policies.

How the Program Works (a hypothesis)

What I’m about to share is a mix of original research, disclosures I got from anonymous sources, and extrapolations from general industry norms. I won’t claim it’s “exactly” how the verified program works, but I’d be willing to bet it matches up well with what most verified account managers would admit if subpoenaed and called to testify under oath. (A scenario I admit is highly unlikely at present.)

Step one: Someone at Substack reaches out to the management team of a celebrity/Notable that isn’t here yet. A dialogue begins, leading to a firm commit that the named principal will “join” Substack (with their team managing the account day-to-day).

Step Two: A coordinated launch, with linked accounts and verification via whatever third party platform the team and Substack agree is best (could be Insta, Linkedin, or X). The badge tracks the fact that this person was verified off-platform, not that Substack verified them directly; this distinction is, in practice, highly irrelevant, since a ton of backend coordination has already happened.

Step Three: Once the account is live and ongoing, the team and the principal work together to handle usage and posting duties. The specifics of who writes what, who replies to messages (both publicly and in DMs), who monitors metrics, et cetera, is all gonna vary by team. I am not alleging that the principals affirmatively don’t write their own content. I am stating that content is probably a team effort, and might be transcribed. All of which is, again, super common and industry standard.

Step Four: Really more of a fork in the road moment. Fork Path A, let’s call it, is that I’d say anywhere from 60% or more of these verified accounts will, after around 6 months, stop being meaningfully active on the platform. That will vary by person, but the pattern is super durable at this point, and predates the verified program. Path B is something like “going commercial,” where the account turns on paid subscriptions and gets a white/orange badge instead of gray/silver. Path C is turning all badges off, like Charli XCX, which is super rare. (Notably, Paths A and C can intersect.)

The strongest framing question remaining might be: who at Substack is telling intermediaries not to discuss it? And the honest answer could be something like “it’s just understood.” Most of these account managers are already under contracts not to disclose their status as the “real” runner of a celebrity’s socials. They’re like a ghostwriter, an assistant, and confidant all in one. It would be bad optics if they were too open about who their clients are or what their operations look like. The verified badges just get folded into a larger code of silence the industry follows.

Closing

I’ll say the boilerplate: “Substack did not substantively respond by the deadline,” not “Substack refused to comment.” Silence isn’t technically a refusal until the subject(s) say so explicitly, even if reasonable minds can make fair inferences. Like we looked at, maybe the pitchforks were coming for celebrity monetization, and a quiet checkmark felt safer than a public fight. But we gotta wonder why. And whether a platform that promised to put writers in charge should even get to settle that argument in the dark.

I gave them a week and every opening to explain. I think decent answers are possible, and I gave you some indications of what those would look like. But Substack still haven’t given us that. Silence like this? I think it’s more than caution, or uncertainty. I think it’s a strategy. And I’ll leave you to decide what you think of that strategy.

Secondly, one time donations are also availaible via Cash App ($Edokwin), PayPal, and/or Venmo (@Edokwin).

Thirdly, anonymous donations and alternative subscriptions are available via at Buy Me a Coffee!

Last but not least, you can schedule a meeting with me via Setmore. If you need my professional advice, or want to work with me on a project, that’s your best bet.

Thanks as always. Hit me up in the comments with smart replies!

See ya next time.

References / Citations

Substack. 2022. “Paying Readers, Not Substack, Decide Who Gets a Badge.” Substack On Official Blog, November 15, 2022.

On Substack

Introducing the Substack Bestseller badge

Today we’re introducing a way to see, at just a glance, whether or not someone is a top-performing writer on Substack: Bestseller badges…

Read more

4 years ago · 346 likes · 547 comments

Quē, Quinn. 2024. “You Don’t Know Jack (or Elon) About Twitter Verification: The Ultimate Deep Dive into the Bluecheck Program on Twitter/X, Its History, Raison d’Etre, and Popular Misconceptions.” Edokwin Editorial, February 7, 2024. https://edokwin.substack.com/p/you-dont-know-jack-or-elon-about.

Quē, Quinn. 2025. “Substack’s Secret Verified Badges: An Exposé.” Edokwin Editorial, November 19, 2025. https://edokwin.substack.com/p/substacks-secret-verified-badges.

Greco, Lachrista. 2025. “Should Celebrities Be on Substack?” Lachrista, Substack.

Rage & Softness with Lachrista Greco

Should Celebrities be on Substack?

If you would like to pay for a subscription, but don’t want to do so through Substack, you can pay via my Venmo or PayPal. Also, if you’d like a paid subscription, but can’t afford it, please email me and I’ll add you—no questions asked…

Read more

a year ago · 23 likes · 9 comments · Lachrista Greco

Kearns, Kim. 2025. “Glennon Doyle’s Departure and What It Means for Writers.” Kim Kearns Editorial, Substack.

Unshattered Sobriety

Glennon Doyle's Departure

I was truly disheartened to receive Glennon’s newsletter this afternoon saying that she’d decided to leave Substack, after being on it for less than a week. She apologized for disappointing her community, but she said that Substack didn’t feel safe for her afterall…

Read more

a year ago · 1690 likes · 301 comments · Kimberly Kearns

Shilling, Vicky. 2025. “What Glennon Doyle’s Time on Substack Teaches Us About Self-Trust.” Vicky Shilling, Substack.

The Business of Self-Trust

What Glennon Doyle’s time on Substack teaches us about self-trust

Last month Glennon Doyle …

Read more

a year ago · 6 likes · 3 comments · Vicky Shilling

McKenzie, Hamish. 2025. “Inside Substack’s Writers’ Civil War.” Interview by Janice Min and Peter Kiefer. The Ankler / Like & Subscribe, June 11, 2025. https://theankler.com/p/inside-substack-writers-civil-war

Taibbi, Matt. 2021. “Meet the Unsuccessfully Censored.” Racket News, March 2021.

Racket News

Meet the (Unsuccessfully) Censored: Jesse Singal and Katie Herzog

On almost a daily basis now, a high-profile figure in the media business is fired or resigns under pressure, often after falling afoul of staff for behavioral or political reasons. The most recent episode involved 27-year-old Alexi McCammond, who this week…

Read more

6 years ago · 180 likes · Matt Taibbi

FOX 29 News. 2022. “Twitter’s Blue Check History: From Anti-Impersonation Tool to Paid Verification.” FOX 29 Philadelphia, November 2, 2022. https://www.fox29.com/news/twitters-blue-check-history-verification-system.

Caplan, Robyn. 2024. “A Working History of the Verified Internet.” Research Series, Data & Society. https://datasociety.net/research-library/a-working-history-of-the-verified-internet-part-three-the-celebrity-wars/.

Ryder, Taryn. n.d. “They’re Influential—and Invisible: Inside the High-Stakes World of Celebrity Social Media Managers.” Yahoo Celebrity (via AOL). https://www.aol.com/theyre-influential-invisible-inside-high-120040320.html.

The Globe and Mail. n.d. “Does Harper Really Tweet from His Own Account?” https://www.theglobeandmail.com/news/politics/does-harper-really-tweet-from-his-own-account/article7419989.

Read more

I have an idea about all that. I’m calling it the “Realignment Thesis,” building of the realignment theory first developed by political scientist V. O. Key Jr in his seminal 1955 journal article, "A Theory of Critical Elections." Per Key and now myself, there are rare but critical events (typically elections) where a sharp, durable shift occurs in voter coalitions and party attachments.

Battleground issues, to use a common term of art, have not only expanded from a shortlist of topics like taxes and foreign relations but exploded as (over)heated clashes of values and cultures in virtually every facet of public and private life. Nothing, and no one, is safe from scrutiny. Everything is political, but more pointed and saliently, everything is partisan. Some old matters have flipped sides, whilst others are strange new entrants.

And in the interest of not getting too caught up in bundles or party platforms, I’ve opted to skip the matter of overarching groupings, in favor of just covering them today individually. Whether they were unpolarized to polarized, coding-switched (“left” to “right” or “right to left”), or something else, I’ll be addressing all the biggest flareups. Not just for taxonomy sake, but because I think the changes are themselves further sources of confusion and conflict. It’s my sense that the biggest divides, and their origins, may yet shock some folks. Writing this certainly surprised me.

In the same way two people can be driven to distraction by arguing over the definition of a word, I think different groups have experienced the grand realignments of the 2020s (and related precedents in the 2010s) as a uniquely unmooring series of events. Hopefully doing this analysis will not only make sense of the chaos but help individuals find their way to a politics that works for them. Even if that looks dramatically divergent from the kind they thought they had only 5 to 10 years ago.

A Note on Terminology & Methodology

I am very pointedly putting all of what follows in (sub)factional and ideological terms rather than simply “right” or “left,” or proxies for same. I’ll explain that briefly.

If we were to put this into crass and arguably far too reductive terms, including regional/national party terms like “Democrat vs Republican” or “Labour vs Conservative/Unionist” I think we’d make several mistakes at once. Firstly, it would make the analysis look region-locked, which is super faulty when we know these patterns are happening throughout the Anglosphere and broader sociopolitical entity we call “the West.” Secondly, it bakes in, and buys into, what I think is actually a false binary. Not only the “Myth” (per the Lewis brothers) of “Left and Right,” but of completely clean and coequal splits across all areas of political life.

Another big reason for understanding, and properly contextualizing, the grand realignments is finally admitting to ourselves that whilst we might be able to put ideologies like conservatism, fascism, libertarianism, and populism on one “side”; and then put others such as Greenism, egalitarianism, liberalism, Marxism, and progressivism on another “side”; we cannot ultimately prove that either “side” is itself internally coherent, let alone worthy of the way we reify it reflexively in our rhetoric.

So instead of just reifying, I’ll be using terms that mean specific things. I’ll address liberals as a group that “leans left” (such that this means anything) and supports personal liberty, economic pragmatism, Enlightenment values, and other such things. Conservatives as a group that supports tradition and abhors kicking down Chesterton’s fence without a good reason. Progressives as a vanguard of social change, strong government action/intervention, and social values that make even liberals tug awkwardly on their collar. Populists as a group opposed to elites and elitism, especially when said elites seem feckless, faulty, or fraudulent. Marxists as extremists of “the left” and Fascists as extremists of “the right,” both committed to things that most of the more moderate factions find abhorrent, rightly understood.

I’ll try not to go too granular, like bringing up Greenism (besides the above name-check) or libertarianism (which can itself be mistaken for a political party). But I will make a point of noting that the lean on a given issue is likely to be driven by its association with certain parties in various countries, and the ideology lean of those parties themselves. Let us not forget how sometimes this becomes a genetic fallacy, or a game of hot-potato-slash-musical-chairs, since it mostly just pegs a topic’s polarity to the person or group least afraid to discuss it, and to advocate about it, openly.

I will also be leveraging data from a variety of sources, including original research, and putting it into the terminology matrix I’m referencing rather the language of political parties or strict “right” and “left.” Sources will include Pew Research and its nine categories of respondent, Gallup, PNAS, and many more. The bibliography will be very truncated, and will delineate some claims as original vs paraphrased from pre-existing sources. As ever, I remind folks that I write news articles and essays, not Wikipedia entries. My work is checkable against observable reality, on a factual basis.

Public Health & Medical Institutions

According to polling and surveys, confidence in public health officials like the CDC dropped dramatically after 2020. But notably, there was a sharp partisan lean to that drop. Whereas populists’ trust cratered, and conservatives’ definitely wavered, that of liberals’ only wobbled (with obvious variance by person), and progressives’ remained somewhat stable. The harder edge liberals, ones long skeptical of institutional power, ended up in a place closer to the populists despite starting much further from the mainstream baseline. But most liberals ended up closer to a midway point between populists and progressives. Conservatives ended up strictly closest to populists.

What does all this mean? Well, distrust became “right-coded” (mostly populist, partly conservative, and some arch-liberal/right-liberal energy) in public health, in vaccines or other traditional medicines, and in medical institutions writ large; and thus, trust became “left-coded” (progressive lean, tacit liberal support). I think it’s fair to say the “lean” ultimately populist vs progressive, or anti-elitist vs elitist and collectivist.

I should note here that Marxists didn’t necessarily say much on this topic specifically, except generic collectivist sloganeering about trusting governments—with the subtext of “as long as they’re collectivist,” like the Chinese Communist Party (CCP) or the Putinist Kremlin government, or progressive and/or Marxist aligned governments in the West. Whereas the fascists, such that they had a take, were largely just freeriding, or drafting off of, the populist anti-elite energy to call for questioning institutions and leaders, again strictly where they (the Fascists) saw this as advantageous, especially since they were angling for more power this decade.

But what actually changed on the ground? Well, a decade ago, routine vaccination, pandemic preparedness, and CDC guidance were technocratic non-issues. Post-2020, due to objective mistakes and lies propagated during the COVID pandemic, itself arguably the high watermark for the Era of Progressive Cultural Hegemony, public health officials and institutions beclowned themselves quite badly, as did governments which tried to follow, or even improvisationally expand upon or revise, the guidelines those entities put out. All with massive political blowback.

One notable flashpoint was the international ascendancy of the BLM (Black Lives Matter) movement and the George Floyd protests, which were given sometimes tacit —or sometimes explicit—approval by governments and health experts despite pandemic rules mandating people stay indoors and most especially not form into crowds. The blatant special pleading created a credibility vacuum that populists and others exploited to argue that public health was itself just a political operation.

Another major swing was the way populists, including conspiracy-minded folks like Robert F. Kennedy, Jr and his “MAHA (Make America Healthy Again)” movement, aligned themselves with more arch populists like the Trump administration. Similar patterns occurred overseas, such as with Germany’s AfD, UK’s Reform Party, and Vox in Spain. There is a general sense in the West that being skeptical of elites overall, and towards healthcare elites in particular (including incredulity or contempt directed at physicians), is a sign of freethinking, even arch-liberty philosophy, and a way to thumb one’s nose at progressive orthodoxy. COVID as catalyst, politics as catharsis.

So, to sum up, this is how distrust of public health emerges, and why it’s “right-coded.” Insofar as populism is a movement or ideology of “the right,” this makes sense. But if we look at conservatism, another movement of “the right,” the coding gets more shaky, as that movement believes in tradition, making distrust in institutions less intuitive from them. Still, the fight is principally populists anti-elitists or anti-institutionalists versus progressives, so it makes sense under those terms.

Food and Pharma

A relevant subset of populist skepticism towards elites and institutions shows that while mistrust of “Big Pharma” remains broad, negative ratings among populists and liberals jumped significantly following corporate vaccine mandates, aligning them with historically “left-leaning” consumer critiques. Likewise, populist, liberal, and conservative views on food sourcing, diets, and so on drove a big backlash to industrial and “artificial” food production. Over the last decade, food and pharma criticism moved from progressive health reformers to the center of right-populist health sovereignty movements like MAHA in the States and similar overseas.

The prevailing sense amongst populists, some liberals, and some conservatives is that countries and provinces need more control over their food supplies, as well as more freedom to reject health protocols or recommendations they don’t agree with. Progressives, some liberals, and some conservatives have oscillated on these issues, sometimes arguing from perspectives dubbed “elitist” or “institutionalist” and even “sheep-like,” per critics. Since the populist backlash also has a conspiratorial tint, it is also impossible to fully separate reasonable critiques of under-explained or poorly justified processes from guttural and low-credibility ideas pegged to global skepticism.

It is therefore important to understand that progressives (and their allies) seem to favor a position on these questions that signals high status or respectability, whereas populists (and their allies) prefer one that signals high independence and agency.

Speech Norms & Content Moderation

Polling trends in the 2020s indicates that conservatives, populists, and other “right-aligned” movements believe social media platforms intentionally censor “right-wing” viewpoints, whereas a majority of progressives and Marxists favor stronger platform moderation to curtail what they call “misinformation” and “disinformation.” In the mid-2010s, Silicon Valley was viewed with bipartisan optimism as a driver of economic innovation. Today, tech infrastructure is deeply politicized around debates over free expression versus public safety and informational accuracy, plus epistemic hygiene.

We have clear evidence that progressive activists, bureaucrats, and politicians leaned on platforms to take a stronger stance on content moderation during both the late 2010s and early 2020s, a period some observers call “Peak Woke” or “Woke 1,” and which I refer to as the peak of Progressive Orthodoxy & Hegemony in politics, media, and culture. We also can understand, just ideologically speaking, that progressives share a sense, along with Marxists, that moderation, even censorship, can be a permissible tool used to shape debate or society, rather than inherently invalid.

Relatedly, conflicts over academic inquiry and freedom were erupting at this same time. Likewise, private individuals and public figures were coming under fire for comments they made, whether in private, in public (via social media or news media or creative media, including comedy), and everything in between. Whilst open debate is not the same as moderation, let alone censorship, it is important to note the chilling effect on speech that happened during this period, especially as regards topics that progressives deemed either inherently taboo or only capable of being discussed in ways that aligned with progressive opinions. It was quite the concoction.

As a result, the support for inquiry, speech, moderation, censorship, and consequences, which had pre-existing political valences that skewed more liberal (and “left”), started to look more like populists versus elitists, especially as regards “right-wing” views, ideologies, and entities being stigmatized or suppressed. The liberals were, and remain, firmly split on these questions, often taking them case by case rather than adopting a blanket position; or in some cases, tried to upright a beleaguered position that opted for universalism and neutrality over rules that seemed overly favorable to one “side” of a given speech question or content debate.

It is also abundantly clear, contra to occasional misconceptions on this score, that conservatives (correctly defined and understood) have never taken a liberal view of speech or content, often preferring some level of moderation themselves. Likewise with fascists, but far worse, as the fascists maintain a mirror view to the Marxists on censorship, thinking it entirely permissible, and even a platonic good, when it serves them. In other words, the progressives and conservatives support some level of moderation, whereas the fascists and Marxists support censorship—similar impulse, just a more extreme execution. Only the liberals and populists resist both, kinda sorta.

So with all this, we must conclude that the conflict over speech and content is less a clean case of a “left” faction vs “right” one, but more of a free-for-all. The liberals want consistent rules, as well as pragmatism. The populists might support that, but mostly just want institutions to stop being elitists about who gets to speak (a sentiment which can sometimes be taken to ridiculous extremes). That some folks want(ed) to leverage the supposed “right-coding” of speech for political gains is becoming inarguably clear as populists, conservatives, and even fascists gain political power.

Transgender Issues vs…

One of maybe the top three areas where the realignment thesis is most salient and edifying would be trans issues, and the conflicts over them. Some trans persons, as well as their allies in the broader gender-affirmer movement, have made pushes to not only normalize transition as a (mostly) administrative matter but to make others agree to their personal declarations, and to the administrative renderings. Up to and including compelled speech, access to sex-segregated spaces, and denial of scientific findings (both stable, long-standing ones and emerging developments).

I say all the above as descriptive realities. Just as I also state, without flinching, that many gender-critical people (sometimes called TERFs) are bullies and bigots. I remain firmly unaligned on trans issues. The fact that I would even be pegged as “picking a side” by accurately describing the actions of activists, academics, physicians, and political entities is itself another part of the problem. Everyone can, and should, call balls and strikes on these matters without such aspersions cast on their motives.

What makes this all so much more complicated is that the fight over gender-affirmation vs gender-criticism is driven by competing coalitions rather than clean factions, whilst trans persons themselves make up part of a larger—albeit increasingly fractured—sexual minority coalition called the LGBT community. I will try to unpack both in as few words as I can, as I don’t want to overcomplicate this any further.

The LGB part of the aforementioned, larger acronym stands for “Lesbians, Gays, and Bisexuals.” All of these are sexual orientations, not gender identities, which is where the trouble begins. But for decades, trans was considered an important part of the LGBT community, if for no other reason than the fact that many trans persons once thought themselves to be one of the other letters, or still did even after transition. And of course, all faced discrimination from the overarching “straight” (heterosexual and cisgender) population, especially from conservatives and fascists.

But as LGB people gained more acceptance, and rights (such as marriage and adoption), most political groups softened their stances on the broader community—even conservatives. Some LGB people themselves, particularly a vocal subset of what’s called the “assimilationist” wing of the LGBT coalition, even argued for political conservatism—just with sexual orientation carveouts. The trans persons, who made up a disproportionate share of the “liberationist” wing of the coalition, were firmly entrenched in progressive, and even Marxist, thought—and did not accept this.

If all that weren’t enough, there had long been a split within feminism (itself another “left” ideology) over LGBT acceptance in general and gender-affirmation in particular. Many feminists took the progressive view that LGB people, most obviously and especially lesbians, deserved equal rights and equal recognition in the mainstream. Other feminists, ones who would become the group that made up the first TERFs (trans exclusionary radical feminists), took an uncertain view of the LGBT community overall and a dim view of trans in specific. For them, sex-segregation for females was non-negotiable, and they knew trans persons would eventually want, then demand, such a negotiation (really, a total capitulation, in the TERFs’ view).

With all that in mind, the current conflict must therefore be understood as a less straightforward case of “pro-trans leftists” versus “anti-trans right-wingers” and more as an omnidirectional case of trans persons (and crucially not all trans people) versus LGB people vs feminists versus conservatives, with plenty of room for internecine squabbles and horse-trading along the way. This is why the gender-criticals, tho ostensibly a coalition of conservatives, anti-progressive populists, freedom-loving liberals, are fundamentally carrying the banner of a feminist cause.

And likewise the gender-affirmers, tho heavily steeped in a weird hodgepodge of liberal, feminist, progressive, and/or Marxist thought, aren’t necessarily any of those at their absolute root (tho calling them “Marxist-progressives” wouldn’t be a stretch). That they happened to gain massive purchase at the height of Progressive Orthodoxy is sometimes leveraged to flatten all these distinctions, framing trans issues as simply “progressives vs conservatives,” but that’s simply not the whole story. Not at all.

As to the specifics, it appears that a majority of Americans support general anti-discrimination protections (including for trans persons), but want these paired with restricting gender-affirming transitions for minors and limiting sports participation by birth sex—with a wide partisan gap on the latter two proposals. (Overseas findings varied.) So a topic that was functionally invisible from mainstream national politics at the start of the millennium, and was still so, well into the early 2010s, had now rocketed to the front lines of legislative battles and international messaging within the last ten years.

What I’ll say to cap this section off is that, whilst trans issues are certainly more than a fight between two groups of feminists, or two subsets of the LGBT community, it’s not immediately clear that this dispute is essential to more than those groups. Populists, some liberals and conservatives adopt anti-trans stances mostly because gender-criticals did a good PR campaign talking them into it (often avoiding questions around male sex-segregated spaces, which feminists typically oppose). In turn, the pro-trans side takes their marching orders from gender-affirmers, who similarly sand off the rough edges of that position. Very few people are honest or consistent on this one.

Marriage Equal Under Fire

A related (sub)topic is the issue of marriage equality. Social liberalization on issues like gay rights has experienced a distinct “right-coalition” retreat, driven by conservative resistance to trans issues, and populist+liberal+feminist resistance to progressive orthodoxy around LGBT issues more broadly. The key takeaway is that the “populist right” didn’t just absorb socially conservative values, nor feminist values; on key social issues, its core base actively reversed earlier liberalizing trajectories, seeming to indicate that support for LGBT peoples was always more fragile than progressives and liberals realized, thus leading them to take it for granted until it was too late.

Protectionist Trade Policy & Tariffs

Protectionism, tariffs, and opposition to global trade (or against neoliberalism) was historically a pillar of labor-union “leftists” or socialists (which naturally included many Marxists). But it was, and is, also a form of populism, since it was anti-elite, specifically opposed to the global(ist) elite. The emergence of “right-”populism inverted this dynamic, transforming what were once ostensibly “hard left” policies into “hard right” ones, and making parties like the GOP in the USA or Reform in the UK into the primary champions of industrial protectionism. Populism works in funny ways.

The Marxists, for their part, have since opted to abandon protectionist rhetoric in favor of internationalist utopian appeals, especially as they push a form of campism that casts the West (the 1st camp) as inherently bad and its enemies (2nd and/or 3rd camps) as inherently good. Progressives, who have been known to take the Marxists’ lead, tend to echo these sentiments. Whereas liberals remain split on these matters; and the conservatives find themselves increasingly forced to agree, however uncomfortably, with the populists for coalition management reasons.

Now let’s add in the conflicts over the evolving voting patterns of working class people (blue-collar workers, non-university graduated youth, and so on). Data from Pew confirms that voters without a four-year college degree favor populist or conservative political parties by around 15 percentage points, representing a historic high. This whilst university-educated individuals favor parties that lean liberal, progressive, or Marxist. Where once working class and “under-educated” voters were central foundations of “leftist” coalitions, such folks have heavily migrated to the “right-wing” parties over social and economic issues—leaving the former flummoxed.

Skepticism of Military & Police

Due again to a populist uprising, favorable views of the national and local police forces roughly were cut in half amongst “right-leaning” groups from the mid-2010s to the mid-2020s. Skepticism of, or support for, police nominally rose within liberal cohorts, with progressives flat or dipping slightly from historic baselines, with major dips due to BLM and major spikes due to events like populist riots in the USA and Europe (January 6th, migration protests, et al). Marxists’ and fascists’ views on police largely tracked which parties were in office, rather than current events.

A more pronounced and durable finding was that support for the military has cratered, especially amongst populists, liberals, and even some conservatives. This too seems to be tracking a populist sense that wars are something which benefit elites at the expense of the common man. Progressives remain staunchly anti-war, as do Marxists. Fascists occupy an awkward liminal space where they posture as nominally anti-war but still edge into some saber-rattling rhetoric against disfavored groups.

Specific events or drivers include prosecutions of political figures, such as current or former heads of state, as well as several successive or concurrent foreign wars and military actions—plus military aid—that either ended unsuccessfully or remain ongoing. Populists feel especially besieged by the sense that police are weaponized against them or their allies, whilst wars are conducted for reasons they view as invalid.

Conservatism still predicts support for police and military, just as progressivism predicts skepticism towards same, but the strongest predictor remains the political makeup of the federal and local governments themselves. Parties viewed as favored by one group will be treated as running criminal justice and defense better than those led opponents.

Relatedly, support for international security alliances such as NATO, and strategic regional partnerships like the various “special relationships” between nations, have taken massive beating due to populist-driven skepticism. (I’ll cover that from the perspective of a singular nation in the Levant in more detail elsewhere below.) The sense, again, is a resurgence of populist, conservative, and even fascistic nationalism over internationalism, with international relations and foreign affairs often framed as zero sum distractions versus domestic policy. Progressives, liberals, and Marxists, ever the internationalists, are thus the dominant face of anti-nationalist sentiment.

What changed in the last 10 years? Hawkish internationalism and defense of transatlantic security were traditional tenets of post-WWII Conservative, pro-military, and proactive foreign policy. The shift toward populist nationalism, non-interventionism, and “our country first” style thinking has turned strategic alliances into partisan issues, with progressives and liberals as the primary defenders of NATO expansion and global democratic alliances or regional partnerships, even arms deals.

Support for Israel & Jews

Tracking highlights a generational and partisan fracture: while older establishment voices remain supportive of Israel, the overall progressive movement and its net sympathy for Palestinians exceeded net sympathy for Israelis, and even domestic Jews, among younger cohorts…for the first time in the mid-2020s. Once a reliable bipartisan issue, Middle East diplomacy became a complex wedge topic that fractures internal party coalitions. Populists remain openly skeptical, often for nationalist and non-interventionist reasons. Fascists and Marxists are critical, often in explicitly ethnic or racialized terms, viewing Israel and Israelis as manipulative, oppressive, and uniquely evil on the world stage. Many local Jews feel the knock-on effects of this.

Conservatives, rightly understood—no pun intended—remain stalwart in their support for Israel and Jews, as do a vocal minority of liberals. (The broader liberal community is heavily mixed and divided.) Progressives, taking their cues from Marxists, view Israel as a uniquely evil country, and increasingly view even domestic Jews with skepticism. Age and party affiliation are again the two strongest predictors here, with older groups that lean conservative or classically liberal viewing both antizionism and antisemitism as nonstarters, whereas younger groups that lean populist, progressive, Marxist, or fascist viewing Israel and Jews with deep suspicion-cum-contempt.

A meta common theme has emerged in these conflicts, which is the idea that supporting Israel (the country or its dominant government) is old-fashioned, whereas being critical of it is both trendy and a sign of freethought or enlightenment. Support or opposition towards Jews as a group is an unfortunate but overdetermined entailment and knock-on effect of these debates, as the vast majority of Jews are Zionists and view Zionism as essentially just the right of Israel to exist, rather than a defense of its current government. Conservatives, especially older ones, tend to support this for geopolitical and cultural reasons. Progressives tend to oppose it as hoary and “antiwoke.” Populists tend to view it as divided focus or divided loyalty.

The longstanding hatred of Israel on the parts of Marxists and fascists are mirrored horseshoes of extremist thought. Where fascists consider antisemitism and antizionism just standard facets of their politics, Marxists view them as necessary entailments of broader campist commitments that center anti-Western, anti-capitalist, and anti-JudeoChristian values, including an unintuitive but longstanding “Red-Green alliance” between Marxists (the “Reds”) and Islamists (the “Greens”). To the Marxist, support for Israel is tantamount to support for Western, capitalist chauvinism (the “1st camp”) and even can count as nationalist ethnic supremacism. Whereas support for Islamism functions as an internationalist movement towards harmony and “praxis.”

Because support for Israel, and strict opposition to antisemitism, have become “right-coded,” many progressives and liberals also feel increasingly uncomfortable pushing back on the increasingly mainstream antizionist arguments. This creates an appearance or phenomenon of “tacit support” that emboldens progressives and Marxists in more “left-leaning” parties. Likewise, the younger populists and fascists in more “right-leaning” parties take cues from this mainstreaming, and often build alliances “across the aisle” on the subjects of Israel and antizionism.

What’s Really Going On

Suffice to say, the meta story of these grand realignments is more nuanced than “left vs right,” or progressives vs populists+conservatives—tho the latter gets closer to the truth. Increasingly, specific issues are taking on partisan valences strictly due to the retreat of one side, or the perceived co-opting/”hijacking” of an issue. A common pattern is the semi-evergreen phenomenon of an issue becoming “conservative” (i.e. traditional) after it’s been mainstream for long enough, but a surprising polarity can arise from populists adopting an issue and progressives reflexively retreating from it.

Likewise, in the case of things like military partnerships, neoliberal trade policies, and even institutionalism, liberals and progressives find themselves in the awkward position of being holdout defenders of what were previously bipartisan positions, due to populist divergence and dissent. In some ways, the biggest predictor of a realignment is not support from a more moderate ideology like liberalism or conservatism, but rather from a more ardent wing like the progressives, populists, Marxists, or fascists. Once a topic is viewed as “toxic” due to its association with extremists, the moderates of one or both sides tend to retreat.

Throughout all of this, of course, the role of political activists, politicians, and parties shouldn’t be overlooked. Whether the GOP, the Dems, or the DSAs in the USA; a revolving door of Prime Ministers in the UK; surging populists or Fascists in Europe; or Marxist and theocratic leaders in nations hostile to the West… it’s clear that many individuals imbibe politics as a bunch of noisy signals from persons and/or groups they trust implicitly, not as discreet ideological understandings. This too creates much opportunity for weird alliances and changes, especially with figures like Donald Trump or Zohran Mamdani, who blur many pre-existing ideological lines to build support.

Close

I think these realignments will only continue and intensify as times goes on. The only thing that the various interested and/or distressed observers and participants can do is try to research the various policy areas and determine what positions, politicians, and parties, which make the most sense to them. This may then yield counterintuitive answers, from split ticket voting to full party shifts.

Gone are the days when people could say that being “left-wing” meant a union of working class and upper class liberals plus progressives who largely agreed on the same core economic and cultural commitments. Similarly, the era of conservatives having a monopoly on “right-wing” thought is long over, with populists and even fascists making heavy gains—ones which restructure the whole ecosystem.

And tho the Marxists and fascists both stand as odious forces in politics, they’ve achieved massive, entryist strides that warrant careful monitoring and vigilance. But of course that vigilance need not necessarily mean running blindly into the arms of an opposition party, since both “sides” have their own competing extremists. Often, it means calling out “both sides” unapologetically, even if that codes as disloyalty.

Secondly, one time donations are also availaible via Cash App ($Edokwin), PayPal, and/or Venmo (@Edokwin).

Thirdly, anonymous donations and alternative subscriptions are available via at Buy Me a Coffee!

Last but not least, you can schedule a meeting with me via Setmore. If you need my professional advice, or want to work with me on a project, that’s your best bet.

Thanks as always. Hit me up in the comments with smart replies!

See ya next time.

Comprehensive Bibliography

Methodological Note: This synthesis integrates survey tracking data, academic frameworks, legal filings, and journalism from 2011 through 2026. Please note: Portions of this analysis reflect original research, synthesis, and extrapolations based on the empirical data points gathered across these sources.

Academic & Political Science Literature

• Barber, Michael, and Charles Pope. “A Comparative Test of Political Cues: Partisanship vs. Ideology.” American Political Science Review (APSR). (Used for cue-taking models on tariffs and protectionism).

• Drutman, Lee. Breaking the Two-Party Doom Loop: The Case for Multiparty Democracy in America. Oxford University Press.

• Grossmann, Matt, and David A. Hopkins. Asymmetric Politics: Ideological Republicans and Group Interest Democrats. Oxford University Press. (Used for asymmetric party sorting baselines).

• Hopkins, David A. Red State, Blue State, Rich State, Poor State: Nationalization of American Politics. University of Chicago Press.

• Key, V. O., Jr. (1955). A Theory of Critical Elections. The Journal of Politics, 17(1), 3–18. https://doi.org/10.2307/2126401

• University of Utah Political Trends Panel. (2019 longitudinal study on voter cue-taking and party identification).

Public Opinion Research & Polling Organizations

• Pew Research Center: Comprehensive tracking (2011–2026) across key metrics:

  • Coalitions & Elections: “Behind Trump’s 2024 Victory,” “Voting Patterns in Recent Elections,” Midterm Preferences (2026).

  • Institutional Trust & Policy: Public trust in science, CDC, FBI/DOJ, election confidence, executive power, and AI sentiment (2021–2026).

  • Foreign & Social Policy: Foreign trade (2015 baseline, 2026 tariffs), Russia threat perception (2017 baseline vs. 2026), youth trans restrictions (2022–2025), and religious identification.

• Gallup Polls:

  • Longitudinal Social & Economic Tracking: Same-sex marriage & LGBTQ moral acceptability (2021–2026), marijuana legalization, K–12 education satisfaction, energy preferences (solar/wind vs. nuclear), corporate trust, and union approval (2001–2026).

  • Foreign Policy & Institutions: Israel/Palestine sympathy and two-state approval (2001–2026), immigration views (2024–2026), and confidence in police.

• The Chicago Council on Global Affairs: Annual surveys (2018–2025/2026) covering Ukraine military aid, anti-restrictionist free trade stances among Democrats, foreign aid support, and immigration threat perceptions.

• PRRI (Public Religion Research Institute): 2025 Youth & Gender Ideology Reports (ideological vs. party ID splits among 18–29-year-olds).

• KFF (Kaiser Family Foundation): Vaccine trust and public health agency sentiment (2023–2026).

• AP-NORC Center for Public Affairs Research: Higher education DEI program favorability (2025) and immigration tracking.

• Quinnipiac University Polls: Federal equity stakes, weaponized use of force/war powers (2025–2026), and spending bill attitudes.

• YouGov / Economist: “Abolish ICE” and law enforcement enforcement sentiment (2026).

Specialized & Advocacy Polling

• Echelon Insights: Foreign aid perception study (commissioned by the Rockefeller Foundation, June 2026) and state-level AI regulation survey (commissioned for Common Sense Media).

• RMG Research / Napolitan News Service: Free speech and media employment consequences post-political violence (Sept 2025).

• Data for Progress & Heatmap: Republican voter economic populism, healthcare, and local infrastructure (data center) attitudes.

• Young Men Research Initiative (YMRI) / Catalist: Post-2024 general election non-white and young male voter shift analyses.

Legal, Policy, & Legislative Records

Judicial Rulings & Statutory Actions

• U.S. Supreme Court Rulings:

  • Learning Resources v. Trump (2026) – Striking down IEEPA tariffs under executive authority.

  • SFFA v. Harvard (2023) – Affirmative action limits impacting corporate DEI.

  • Loper Bright Enterprises v. Raimondo (2024) – Administrative state and regulatory deference.

  • United States v. Skrmetti (2025), West Virginia v. B.P.J. (2026), and Little v. Hecox (2026) – State restrictions on youth transition care and school athletics.

  • Trump v. Cook (2026) – Federal Reserve leadership removal protections.

Legal & Regulatory Analysis Outlets

• Lawfare: Legislative tracking of presidential war powers, use of lethal force, and executive authority votes (2025–2026).

• Jones Day & Snell & Wilmer: Legal summaries on Section 122 surcharges, Section 301 duties, and Court of International Trade tariff rulings.

• Hogan Lovells: Federal Trade Commission (FTC) v. Meta antitrust litigation monitoring.

• JURIST & Texas Tribune: State and federal court tracking on sports and healthcare access.

Media Reporting & Analytical Secondary Sources

Public Policy & Economic Analysis Outlets

• City Journal (Manhattan Institute): Analysis of NBER corporate communication data regarding S&P 500 Twitter/X language and ESG/DEI trends (2017–2023).

• Niskanen Center: Analysis of Republican trade stances and asymmetrical party polarization models.

• Advisor Perspectives & Richmond Fed Index: Michigan Consumer Sentiment partisan gap tracking across administrations.

• GovTech: State vs. federal artificial intelligence regulation policy analyses.

• VoteHub: Post-2024/2025 demographic election modeling and thermostatic turnout analysis.

Journalism & News Reporting

• Associated Press (AP): Coverage of state transition care bans, immigration trends, Democratic candidate positioning in 2025–2026 elections, and Senate foreign military sales votes.

• Axios: Public approval tracking of labor unions (2024–2026) and big business threat perception.

• Reuters / Ipsos: Middle East military action approval, corporate taxation polling, and presidential approval among non-white demographics.

For t…

Read more

I made a tweet, declaring eight men as uniquely qualified, cogent, and calibrated on the questions of AI in general and AGI (artificial general intelligence) in particular. I implored people to look these men and their writings up for themselves, and stress test any of the claims they were making. I believed then, and still mostly believe now, that they held up well against a sea of nonsense that gets said in this domain.

Well, time to apply the method. Today, I’ll be re-investigating them for you. I’ll look at what they said, what counts as having “held up,” what fell short, and why they are (mostly) all still the best people to follow on the questions of AI as a normal technology rather than a millenarian cult exercise. Let’s get started.

Rodney Brooks

An Australian roboticist and computer scientist, Brooks is former director of the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) and co-founder of iRobot (makers of the Roomba) and Rethink Robotics. In the late 1980s, he introduced behavior-based robotics and the Subsumption Architecture—a paradigm shift that rejected the idea that robots needed complex, centralized internal models of the world to act, instead arguing that "the world is its own best model." If this sounds similar to my own thinking around Moorean realism and philosophical rationality, that’s already a bit of a hint as to why I like his thinking and how he came on my radar.

During the Generative AI and LLM craze that began in the early 2020s and hit a fever pitch in 2024, one which led to public discourse that was dominated by aggressive timelines for Artificial General Intelligence (AGI) and/or AI Doom scenarios, Brooks emerged as a leading realist—a useful, pragmatic counterweight to hypetrains and apocalyptic rhetoric coming out of Silicon Valley and beyond. And we needed him.

Brooks coined the acronym FOBAWTPALSL (”Fear Of Being A Wimpy Techno-Pessimist And Looking Stupid Later”) to describe the herd mentality driving tech CEOs and investors. While mainstream narratives projected human-level intelligence and ubiquitous humanoid labor within 3–5 years, Brooks argued that the industry was mistaking raw scaling of statistical pattern matching for true reasoning. He leaned into his prior wins—the aforementioned revolutionary innovations in robotics—to make strong bets against the nonsense we were all seeing. And he had a tool to keep himself honest, whilst also logging his wisdom for posterity: an annual Scorecard.

Since 2018, Brooks has tracked and updated a public Predictions Scorecard, assigning dates for when specific AI and robotics breakthroughs will—and will not—happen. He discusses AI, cars, space travel, and other domains of interest and relevance. He scores himself annually, and updates as needed, for the nearest term predictions, whilst logging new ones and holding firm on the existing bets not yet cashed out.

He was right about robotaxis, SpaceX bluster, and in-home care robots. He was wrong on electric vehicles only because he erred on the side of tempered optimism in an environment that was even more embarrassingly bullish (the Tech Bro Elite) and ended up logging even worse predictions than his—a sort of Nate Silver moment, where his moral victory was being wrong for the right reasons, in an area where most people couldn’t even see straight. He still has live predictions for LLMs that need time.

By far the best thing I read from him initially, when I first discovered him, was his “Seven Deadly Sins of AI Predictions“ article, which laid out his core framework all the way back in 2017 and has aged like wine since. Building off the thinking of men like Roy Amara and Marvin Minsky, amongst many others, Brooks explained that the core “sins” of AI discourse are: overestimation & underestimation (extreme optimism vs extreme pessimism basically); magical thinking or imagining sci-fi in place of real/normal technology; confusing performance of one skill for general competence in multiple domains; leveraging strategic ambiguity via “suitcase words” that carry multiple meanings, making clarity and accuracy difficult-to-impossible; taking exponential growth as a given, and refusing to grant friction or constraints.

He listed seven points, but close reading would reveal that points #2 & #6 are directionally the same critique (magical/fantastical imaginations run wild), and that #5 & #7 likewise point to similar issues around taking runaway development for granted. Still, a worthwhile read and a great corrective to nonsense. As LLM scaling curves encounter diminished returns and economic costs push companies toward specialized agentic architectures rather than pure brute-force training, Brooks' insistence on hybrid architectures and safety boundaries has been largely vindicated.

François Chollet

Chollet is a French AI researcher, software engineer, and one of the most prominent pragmatic voices in artificial intelligence. He created of Keras, the deep learning API that democratized AI development in the mid-2010s; authored the book Deep Learning with Python, and worked as a Senior Staff Engineer at Google from 2015 to 2024, including during the height of the first AI boom, which is when he hit my desk.

During the peak hype of that boom, Chollet became one of the Top 3, if not just straight THE, leading intellectual counterweight to people like Sam Altman, Dario Amodei, and Silicon Valley’s “scaling hypothesis”—the belief that making models bigger, with more compute and data, will naturally produce Artificial General Intelligence (AGI). He set out to prove the opposite, with mixed results.

In 2019, Chollet authored a landmark paper, “On the Measure of Intelligence,” proposing the Abstraction and Reasoning Corpus (ARC) challenge. Unlike traditional benchmarks that test memorized knowledge or pattern matching, ARC would test skill-acquisition efficiency on visual logic puzzles an AI has never seen before.

Fastforward to mid-2024, and Chollet launched the ARC Prize (a $1,000,000 challenge co-created with Mike Knoop) to push the AI industry away from LLM brute-forcing. While top LLMs were scoring north of 90% on human university-style exams, they stagnated around ~20%–30% on ARC, proving Chollet’s point that memorization isn't real reasoning. He did a tour of podcasts and blogs, making bold predictions that AIs from leading firms wouldn’t solve ARC—hell, wouldn’t even reach scores north of 60%—for years to come. The results seemed to back him up, until they sorta didn’t.

Whilst his core argument, that LLMs are interpolative database lookup engines rather than autonomous reasoning systems, may yet be sound, something unexpected happened in December 2024. OpenAI unveiled o3, a new and custom model. Under high test-time rules (allowing the model to “think” and try thousands of internal program hypotheses before answering), o3 achieved an 87.5% score on the semi-private ARC-AGI-1 evaluation set. For context, human baseline performance on ARC sits around 85%. To many AI advocates and scaling-hypothesis believers, this was seen as a definitive “checkmate” moment against Chollet’s thesis.

Yet he maintained that true AGI requires synthesis on the fly, and that OpenAI had to burn hundreds of thousands of dollars and use intensive, specially designed hardware, plus pour massive computing power (“compute”) into something that scaling alone cannot achieve. He compared the relatively breezy effort a human would apply to get a 85% without any assistance vs what o3 required, and said these weren’t equivalent. Later ARCs have fared similarly, with the AI firms getting low scores when testing their main, older models versus higher scores when they use their newest and best.

After leaving Google to co-found Ndea, an independent AI research startup, Chollet wanted to focus directly on building novel architectures centered on program synthesis, continuous adaptation, and genuine fluid intelligence. He felt that Google’s DeepMind AI division, lead by Demis Hassabis, was at odds with his current philosophy and that he could do more going independent. He also knew that his continued leadership of the ARC would create conflicts of interest if he stayed, given DeepMind’s stake in competing in the annual test, like other AI firms.

I respect Chollet for his unflinching heterodoxy in a field that often runs on groupthink. I also believe he’s fundamentally correct to doubt that the leading AI firms are progressing towards general intelligence. The fact that he’s been a bit overly bearish on benchmarks and timelines is the cost of being willing to buck the herd.

David Deutsch

Deutsch is a pioneering British theoretical physicist at Oxford, recognized as one of the founding fathers of quantum computation for formulating the universal quantum computer, building off the work of proto-computing legends like Alan Turing. Beyond physics, Deutsch is a major philosopher of science whose books (The Fabric of Reality and The Beginning of Infinity) detail a worldview built on Popperian epistemology, optimism, and universal computation. And his distance from the ideology, commercial, and physical bubbles of Bay Area tech bros is a plus rather than a minus.

For decades, going all the way back to the first AI booms and busts of the late twentieth century, Deustch has stood firm on three basic commitments: first, that the concept of “true AI” or AGI is extremely far off, maybe even illusory, given our current limitations with the technology in particular and with understanding cognition in general (the “hard problem of consciousness”); second, that a hypothetical AGI would be unlike anything we’ve ever built with machines, and would not simply come from brute-forcing or compute scaling methods; third, that much of the “alignment” talk is deeply confused and morally/philosophically dubious, since it both begs the question on AI sentience…and yet assumes we can essentially enslave AGIs by rote.

His position is nuanced. I find it, and him, so interesting and so useful precisely because he simultaneously says: what the AI firms of today are doing is definitionally not AGI and cannot become so; and yet, if AGI were possible, we’d have to reckon with it like we reckon with a living organism, which is to say with respect and understanding, not hard coded rules or expectations of blind obedience.

Deutsch’s core value could be summed up as “optimistic rationalism,” or the view that all problems are soluble given knowledge, with the understanding that all knowledge creation or discovery requires creative critique and conjecture. To put all that in layman’s terms, he thinks we can overcome any problem by thinking about it enough, and be defines quality thinking as the kind of thing that uses originality, challenges ourselves and/or our existing ideas, makes testable claims, and survives scrutiny.

He views AGI not as a hardware or data problem, but as a philosophical/software breakthrough that could be far away because we lack a correct theory of thinking. His stance has remained remarkably consistent from his books each decade to his comments during the 2023–2026 Large Language Model hype cycles. He refuses to give timeline projections or spout nonsense about P(Doom) percentages. He suggests that there are multiple, equally possible scenarios on offer: in one, we could make a breakthrough in philosophy of mind that opens up a huge runway for AGI within our lifetime; in another, we could be stuck going nowhere towards AGI for over a century.

Jaron Lanier

A computer scientist, author, and tech pioneer frequently credited with coining and popularizing the term “Virtual Reality” in the 1980s, Lanier is currently the Prime Unifying Scientist at Microsoft. From his perch, he occupies a rare vantage point: he operates inside a major tech empire while serving as one of Silicon Valley’s most incisive philosophical and structural critics. He doesn’t even think “AI” exists!

Okay, that’s gilding the lily just a bit. His foundational view is that “AI” is an umbrella term for technologies which serve as a medium of collaboration—a vast tool for human social expression that is currently obscured by bad economic models and mythologized framing. Sort of like how someone could look at a car or a plane and imagine it to be some kind of large animal moving of its own volition, rather than a tool which humans direct and use in concert with each other. An interesting pivot.

During the height of post-ChatGPT mania in early 2023, as the narrative was dominated by competing camps of hyperbole and extremism around AI Doomerism (existential risk) vs AI Maximalism (AGI as a messianic savior of society), Lanier broke through the noise. In April 2023, his widely influential essay entitled “There Is No A.I.” was the first salvo, followed by key pieces like “How to Picture A.I.” and high-profile lectures. He has talked extensively about what he calls the need to “invert” our understanding of AI from an autonomous entity to a format of tech collaboration.

The prevailing tech narrative frames AI as a newly created “creature,” “agent,” or “pseudo-person” sitting across from us in a chatbox. Lanier flips this upside down: AI is not an entity; it is a lens through which we view a giant, crowd-sourced collaboration of millions of real human beings. My vehicle analog probably undersells the totality of his frame, because it’s probably better to say that AI for Lanier is closer to a road itself, or to what many did (and some still do) view the internet as.

One useful way Lanier’s framework cuts through a lot of noise and self-serving rhetoric is by recognizing the way tech executives actively leverage the “AI as an autonomous agent” myth as a legal shield. If a model hallucinates libel, leaks private data, or hacks a company, AI firms can claim, the AI did it, and that they don’t fully understand how it works, so they’re not liable. It’s quite the sleight of hand.

Most importantly for understanding the Doom vs machine-god narratives, Lanier rejects the idea of a technological Singularity as a religious impulse that erases human history and memory. He has instead argued that "human-centric optimism" is the way forward, especially when paired with appropriate regulations on AI firms as normal technology companies and fair compensation for people affected by their actions.

If we abandon the god-building ideology and treat software as a tool for human collaboration, per Lanier, AI can be safely integrated without destroying creative labor or concentrating economic power into central monopolies or monopsonies. I think he remains one of the Top 5 best people to read on this topic, even amongst my list, precisely because his thinking is so unique and his framework cuts through so much noise to hit what are really the bedrock considerations at play in the AI debate.

Yann LeCun

The founding director of Meta’s FAIR (Fundamental AI Research), Yann André Le Cun (commonly referred to as just “LeCun”) is a dual French-American citizen and widely considered one of the three “Godfathers of AI” alongside Geoffrey Hinton and Yoshua Bengio. Together with the other two men, he became a joint Turing Award recipient in 2018 for their work on deep learning. And tho still friendly, the three diverge sharply.

During the “AI Winter” of the 1990s and early 2000s, when neural network research was widely dismissed as a dead end by mainstream computer science, LeCun, Hinton, and Bengio were part of a tight-knit, insular network of researchers (informally dubbed the “Deep Learning Mafia”). They co-authored papers, shared hardware strategies, and built the groundwork for modern machine learning, culminating in their joint 2018 Turing Award. Yet even by the early 2010s, LeCun was breaking ranks from his peers on the science, the career path, and the public commentary.

Bengio’s journey could be described as a mostly theory-based specialist, musing about AI from the ivory tower of the Université de Montréal and running niche experiments at Mila. Hinton’s was a mix of small startups, managerial makework at Google, and a messy exit from that company to go on a dramatic AI Doom speaking tour. Throughout this same timespan, LeCun was the only one of the three “Godfathers” consistently in the trenches, doing real systems building, hardware/software co-designing, and championing open-source competition.

Tho personally still warm (by all accounts), the men make mutually exclusive public pronouncements about the industry. On platforms like X (formerly Twitter), LeCun is famously combative with AI safety researchers, whilst his direct interactions with Hinton and Bengio tend to carry a tone of gentle, exasperated ribbing among old friends who know each other's arguments inside out. LeCun routinely clarifies that while he thinks his peers’ existential risk models are wildly incorrect and harmful to public policy, he retains deep affection and respect for them as scientists.

A vehement critic of apocalyptic “AI Doom” scenarios, LeCun has been firm about three main counterweights: he soberly reminds us that they’re speculative fearmongering rather than grounded scientific or technical analysis; he frequently noted in the ‘23-’24 cycle that the majority of hands-on AI workers like himself don’t cosign what people like Eliezer Yudkowsky and Sam Altman say; and he wisely notes, as other observers have echoed, that claims of existential risk and calls for government intervention invite regulatory capture by incumbents.

After over a decade at Meta, LeCun left in 2025 to found his own company focused on world-model architectures and human-like artificial intelligence he calls superintelligence. The resultant company Advanced Machine Intelligence Labs (or AMI Labs), is focused on building AI “world models”: systems that learn to understand the physical world’s structure and dynamics rather than just predict text like LLMs.

What I love about LeCun is his unflinching courage to call bullshit, even when doing so puts him in direct (but mostly respectful) conflict with old friends. He’s a hands-on technologist who knows that many of the loudest voices on AI Discourse™ aren’t necessarily the most well versed on the nuts and bolts, let alone the most credible. That matters for judging him. The single most important thing Yann LeCun has been right about is his early-stake and long unwavering conviction that perception, structure, and spatial invariance must precede abstraction. That has implications both for how to build AI software and how to talk about the field in public.

Vishal Misra

Misra is the RKS Family Professor of Computer Science at Columbia University and the Vice Dean of Computing and AI at Columbia Engineering. He is a seasoned computer scientist, internet economics scholar. During the peak hype cycle of LLMs in 2023 and 2024, Misra emerged as a distinct, pragmatic voice cutting through the dual noise of "AI doom" hype and uncritical commercial optimism.

He entered mainstream AI discourse consistently urging for calm, framing AI not as an autonomous threat to humanity, but as a powerful software tool and infrastructure layer. He drew direct parallels to the networking and internet boom of the late 1990s. Like others on my list, he was in the Normal Technology camp.

Recognizing that generative tools were fundamentally an architectural change in computing—not a temporary phase or a simple cheating mechanism—Columbia Engineering created a brand-new administrative role in early 2024: Vice Dean of Computing and AI, appointing Misra as its inaugural head. After taking the reins, he became a leading academic voice on integrating AI into higher education curricula, calling for proactive policy rather than bans or institutional panic.

Misra brought an equally sharp systems-engineering perspective to the broader convo via the “Bayesian Geometry” framework, which he developed alongside researchers Naman Agarwal and Siddhartha R. Dalal. They mathematically posited that AI’s processes map directly onto Bayesian inference, and are both explorable and explicable, rather than some arcane “black box” that many others posit. Per them, AI models are literally tracking hypotheses and updating context in real time, tho not via cognition in a human sense. Something closer to principled probabilistic navigation.

Regarding AI safety, Misra’s stance has remained remarkably consistent, marked by a refusal to endorse existential “AI Doom” claims while holding a firm position on concrete, systemic engineering and policy risks. His core perspective starts by decoupling structural engineering from silly existential doom talk. He rejects AGI existentialism or millenarianism; he views speculative P(Doom) scenarios as unfounded, since they beg the question on self-aware superintelligence capable of rogue autonomy; and he maintains that AIs are fundamentally mathematical function approximators whose mechanics (such as Bayesian updating) are fully explorable.

Instead of framing safety as an anti-extinction challenge, he focuses on practical software and system vulnerabilities like hallucination bounds, data security, model poisoning, and over-reliance on unverified statistical outputs in critical systems such as business, healthcare, or government. Rather than trying to slow down base-model capabilities out of fear, he advocates for using AI tools defensively and proactively in the fight. By deploying automated verification, adversarial testing, and strict API-level sandboxing, we can build resilient software environments with AI, not in spite of it.

Yet Misra has maintained a strict line on the cognition question, which he bundles into his signature metaphor: the “Einstein Test.” It serves as his benchmark for determining whether an AI system is merely an extraordinarily powerful pattern-matcher or a genuine intelligence capable of redrawing conceptual maps. A Bayesian filter can flawlessly navigate an existing map, but it cannot redraw the map. Einstein didn’t just calculate Newtonian mechanics better than his peers; he constructed an entirely new conceptual framework. AIs are still just following the available maps.

In Misra’s view, AI operates inside “Shannon’s terrain” (referring to Claude Shannon’s work on statistical relationships) by compressing known statistics, but it cannot cross the threshold into genuine theory-building. It does the available work, it does not contextualize that work for, or by, itself. To pass the Einstein Test, an AI system cannot simply prove complex theorems by searching through known pathways. It must autonomously generate a new, lower-complexity hypothesis class—a shorter program or novel framework—that explains anomalous data. He maintains that none have done so to date, and I think the available evidence backs him up cleanly.

Overall, Misra’s position remains subtle, nuanced, and precise: AI is a normal technology that we can and should study, develop, integrate, and control with firm guardrails. He reminds us that two things can be true at once: AI can be a normal technology, more analogous to cars or planes than a digital god; and AI can be massively powerful, disruptive, dangerous, and thus in desperate need of people who take it seriously rather than lightly. Indeed, by understanding it, we protect ourselves. In many ways, I think my own view on the tech has done well to honor both of these.

Judea Pearl

Pearl is a computer scientist, statistician, and 2011 Turing Award recipient. He is widely regarded as the foundational architect of modern probabilistic AI and the pioneer of Structural Causal Models (SCMs) and Bayesian Belief Networks, the latter of which it is a visual and mathematical map of how different factors influence one another under conditions of uncertainty, allowing a computer to update its beliefs about unknown variables when new evidence comes in. He is also a rare “double crown” via his dual Fellowship honors at both IEEE and ACM (an achievement lesser than 0.5% of eligible members reach), the two premier global professional associations for computing, engineering, and technology. His foundational contributions span both physical/applied systems and theoretical software architecture.

For years, Pearl’s words stood as the ultimate theoretical roadblock to deep learning hype. When the tech world split into two camps of scaling optimists (who argued that enough data and compute would yield AGI) versus and architectural critics, Pearl emerged as one of the Top 10 most important—if not always most visible—scientifically literate members of the latter camp. He still stands as an important counterweight to the scaling narrative, which itself has largely fringed his side.

His “Ladder of Causation” framework established a strict, mathematically rigorous hierarchy and the boundaries between states within it. At the bottom, which he dubbed Level 1, is observation and association. Entities or tools at Rung 1 can use statistical pattern-matching to produce results, but they cannot produce causal intervention.

Which brings us to Rung 2, the place where doers and “do-operators“ begin to emerge and grow. Pearl uses what he calls “do-calculus” describe the process of manipulating variables and directly altering systems. Doing is forcing something new to happen using a world model or a causal chain of its own, not specific instructions.

Finally, at Rung 3, Pearl says that we can find counterfactual imagination and retrospective reasoning. Here, we can evaluate hypotheticals that aren’t present in our current modeling or which explicitly contradict actual history. We can postulate what would happen in alternate causal chains, or if different actions were taken. This is the foundation of human concepts like regret, blame, credit, and safety analysis.

To Pearl, AIs (especially in the form of LLMs), were just sophisticated curve-fitters stuck at Rung 1. Brilliant at correlation, totally blind to cause and effect (Level 2), and obviously incapable of actions of cognition upstream in Rung 3. Standard deep learning, classical machine learning, and transformer-based models would stay at the bottom of the Ladder indefinitely, because no amount of scaling or computing power could overcome a fundamental deficiency like this. Just like a better fuel can’t make an old car run like a jet plane, or more mud can’t make a mud hut into a stone castle.

Then came the text-scaling explosion, and Pearl did something rare: he updated in public. Pearl acknowledged a critical nuance he had previously set aside: human text itself is saturated with embedded causal reasoning. An AI trained on billions of pages of human explanations is absorbing the distilled causal models of millions of human authors. By operating on text that encodes Rung 2 and Rung 3 reasoning, the system can proxy causal inference without having constructed the causal graph from scratch.

For this reason, our causal AI researcher Pearl had to offer a revision: he places contemporary reinforcement learning in AI at roughly Rung 1.75 (“level one and three-fourths”) to capture a unique paradox that LLMs appear to answer Rung 2 and Rung 3 questions, even though their underlying architecture is purely Rung 1. He still argues, as best I can tell from a 2026 conversation with Sam Harris, that LLMs recycle human-made world models rather than discover new ones, and that explicit causal models are what’s missing. But he’s crucially pragmatic.

And Pearl is far more willing than LeCun or Brooks to take existential risk seriously. In a 2023 Amstat News interview with Dana Mackenzie with, he warned that once real general AI exists, we don’t even have the metaphors to think about controlling it. He avoided some of the dramatic flourishing and doomsaying of the worst AI Doom commentators, but he’s firmly in what some are dubbing the emerging “AI Safety” movement, an ecosystem that runs the gamut from common sense to nonsense. I don’t have to agree with all his priors or how he got there, but I won’t dismiss him. He’s proven himself credible enough that even his iffy calls deserve a fair hearing.

Ted Sanders

Sanders is an AI safety researcher and OpenAI engineer best known for his empirical work on existential risk quantification and long-term AI forecasting. Operating primarily within the AI risk and Effective Altruism (EA) research ecosystems, Sanders has focused on modeling timeline probability distributions for transformative AGI and assessing catastrophic risk thresholds for frontier LLMs. He was also a key contributor and finalist, with Ari Allyn-Feuer, in Open Philanthropy’s AI Worldviews contest.

In 2023, he challenged aggressive AGI timelines by arguing that “Transformative AGI before 2043 was less than 1% likely“ when properly accounting for deployment bottlenecks, real-world data-collection feedback loops, and potential system derailments. He and Allyn-Feuer introduced an applied physics discipline to timeline forecasting with their 023 paper projecting a 0.4% chance of transformative AGI by 2043. Their strength was evaluating what the physical “bar height” in the form of things like power grids, semiconductor manufacturing, capital constraints, and data limits. They were right that physical deployment is vastly harder than software generation. The forecast has proven remarkably durable, much better than rivals’.

Tho he was skeptical about timelines and probabilities, he took the existential risk crowd seriously enough to posit what became known as the “1 in 10,000 Rule“ for risk tolerance thresholds: sparking widespread debate across the technical alignment community, Sanders suggested that a 1 in 10,000 risk of human extinction per major GPT model version/iteration (e.g., GPT-5) could be acceptable when balanced against the massive expected benefits of advanced AI deployment. His position wasn’t that the AI Doom crowd didn’t know what they were talking about, moreso that they weren’t weighing the pros, cons, and actual odds properly, even on their own terms.

For the next several months during the ‘23-’24 AI cycle, Sanders worked on adversarial collaboration regarding AI Risk, including team-based, quantitative risk-modeling efforts aimed at bringing skeptics and x-risk advocates into a single mathematical framework, aiming to reduce dogmatic polarization between “doomers” and “accelerationists.” These produced mixed results, as participants were often unwilling to revise their priors or converge on one framework, even after open dialogue.

As far as Sanders’ current position and status, he continues to advocate for human-centered AI systems, quantitative risk assessments, and rigorous empirical testing over theoretical catastrophe models, working to bridge the gap between technical operations and alignment policy. His skeptical stance on near-term transformative AGI, pegging it to around 2043 or later, has aged into a central baseline for many pragmatic safety researchers and commentators. Despite being firmly in many of the camps I roundly criticize as wrongheaded and foolhardy on AI, Sanders remains one of the Top 5 most rational and informative voices on AI safety, esp his milieus.

Conclusion

I’ve never said, and never will say, that AI is worthless or can’t do remarkable things. I’ve never claimed to be infallible or free from concerns on the topic. But I have always warned that people who talk out their asses, and regularly show their asses, are driving these conversations, with deleterious effects. That’s still true.

When you look across the entire landscape, the eight thinkers I listed represent the highest concentration of signal in the field. The ones who fared best were those whose formal frameworks gave them a clear baseline, but whose commitment to empirical truth allowed them to acknowledge when new events pushed past their expected boundaries. They forced the industry to define its terms, measure its benchmarks, and confront physical reality. In an era dominated by speculative theater, their collective work remains the only sane baseline for evaluating the actual state of machine intelligence, AI industrial activity, and policy debates about same.

I have included a mix of men, including some with backgrounds pegged explicitly to the communities I find least cogent on these topics (like the Effective Altruists, Bayesian Rationalists, and LessWrong forumers). Why? Because too much arrogant dismissal of the Bay Area AI cultists can turn someone into a different kind of pseud: a Gary Marcus style overclaimer who constantly speaks beyond the available data just for the sake of being a contrarian or throwing water on a party. I am not that guy.

None of these men are oracles. What they have in common is that they put their names on claims specific enough to be checked, and in this field that’s rarer than it should be. Several of those claims took damage in the years between when they made them and now. Most held, tho some are bets we still can’t cash out for years or decades. Still, I’d rather follow people who can be wrong in public than prophets who can’t, and that’s the standard that keeps me sane. Method over millenarian madness.

Secondly, one time donations are also availaible via Cash App ($Edokwin), PayPal, and/or Venmo (@Edokwin).

Thirdly, anonymous donations and alternative subscriptions are available via at Buy Me a Coffee!

Last but not least, you can schedule a meeting with me via Setmore. If you need my professional advice, or want to work with me on a project, that’s your best bet.

Thanks as always. Hit me up in the comments with smart replies!

See ya next time.

Bibliography

Rodney Brooks

• “Predictions Scorecard, 2026 January 01,” Jan 1, 2026. https://rodneybrooks.com/predictions-scorecard-2026-january-01/

• “Five new technology predictions for the next decade,” Mar–Apr 2025. https://rodneybrooks.com/five-new-technology-predictions-for-the-next-decade/

• “Why Today’s Humanoids Won’t Learn Dexterity,” 2025. https://rodneybrooks.com/why-todays-humanoids-wont-learn-dexterity/

• “Four Time Scales for Technology Development and Deployment,” Aug 1, 2026. https://rodneybrooks.com/four-time-scales-for-technology-development-and-deployment/

• “My Dated Predictions,” Jan 1, 2018 (the original list). https://rodneybrooks.com/my-dated-predictions/

François Chollet

• Chollet et al., “ARC-AGI-3: A New Challenge for Frontier Agentic Intelligence,” Mar 2026. https://arxiv.org/abs/2603.24621

• Chollet, Knoop, Kamradt, Landers, “ARC Prize 2025: Technical Report,” Jan 15, 2026

• Y Combinator Lightcone, “How François Chollet Is Building A New Path To AGI,” Mar 27, 2026. https://www.castfox.net/podcast/y-combinator-startup-podcast-655823/episode/how-franois-chollet-is-building-a-new-path-to-agi-462201

• Dwarkesh Patel conversation (timeline cut to “five-ish”), via The Neuron’s write-up, undated. https://www.theneuron.ai/explainer-articles/the-clock-is-ticking-ai-skepticfrancois-chollet-slashes-agi-timeline-to-5-years-unveils-new-benchmark-to-prove-it/

• “On the Measure of Intelligence,” 2019 (arXiv:1911.01547).

David Deutsch

• Altman–Deutsch fireside chat, Sept 24, 2025 (Yahoo/Tech coverage). https://tech.yahoo.com/ai/chatgpt/articles/openai-ceo-questions-whether-ultimate-130000206.html

• Sana “Strange Loop” podcast interview (”every AGI is a person”), via Digital Trends, undated. https://es.digitaltrends.com/computadoras/el-padre-de-la-computacion-cuantica-cree-que-agi-sera-una-persona/

• Tim Ferriss Show #662, with Naval Ravikant, Mar 24, 2023. https://tim.blog/2023/03/24/david-deutsch-naval-ravikant/

• The Beginning of Infinity, 2011.

Jaron Lanier

• StarTalk interview with Neil deGrasse Tyson, summer 2026, via TechCentral (Jul 24). https://www.techcentral.ie/there-is-no-ai-just-a-vast-human-collaboration/

• Same episode via Boing Boing, Aug 14, 2026. https://boingboing.net/2026/08/14/jaron-lanier-there-is-no-ai.html

• “Data Dignity and the Inversion of AI,” UC Berkeley talk, Sept 13, 2023. https://cdss.berkeley.edu/video/data-dignity-and-inversion-ai-jaron-lanier

• “There Is No A.I.,” The New Yorker, 2023.

Yann LeCun

• MIT Technology Review interview, Jan 22, 2026. https://www.technologyreview.com/2026/01/22/1131661/yann-lecuns-new-venture-ami-labs/

• RAISE Summit 2026 talk, via French Tech Journal, about July 2026. https://ami.frenchtechjournal.com/news/lecun-at-raise-llms-hit-a-wall-and-europe-can-win-the-next-race

• AMI Labs launch and $1.03B seed, Mar 2026 (secondary). https://quantumzeitgeist.com/world-models-lecun-post-llm/

• “A Path Towards Autonomous Machine Intelligence,” 2022 (OpenReview).

Vishal Misra

• Agarwal, Dalal, Misra, “The Bayesian Geometry of Transformer Attention” (Paper I), Dec 2025. https://arxiv.org/abs/2512.22471

• “Gradient Dynamics of Attention: How Cross-Entropy Sculpts Bayesian Manifolds” (Paper II). https://arxiv.org/pdf/2512.22473v4

• “Geometric Scaling of Bayesian Inference in LLMs” (Paper III). https://arxiv.org/pdf/2512.23752v4

• “Attention Is Bayesian Inference,” Medium, Dec 30, 2025. https://medium.com/@vishalmisra/attention-is-bayesian-inference-578c25db4501

• a16z Podcast, “Columbia CS Professor: Why LLMs Can’t Discover New Science,” Oct 13, 2025. https://a16z.com/podcast/columbia-cs-professor-why-llms-cant-discover-new-science/

• Columbia post, “Language models: flawlessly reasoning or confidently hallucinating?” Oct 16, 2025. https://www.ee.columbia.edu/news/language-models-flawlessly-reasoning-or-confidently-hallucinating

• a16z, “What’s Missing Between LLMs and AGI,” Mar 17, 2026 (summary page). https://stationlm.com/analysis/2f1d5426-51fc-492e-9114-52ef794b2067

Judea Pearl

• Amstat News interview with Dana Mackenzie, Sept 1, 2023. https://magazine.amstat.org/blog/2023/09/01/judeapearl/

• Making Sense #453, Jan 16, 2026 (paywalled; the AI segment is also a free ~25-minute excerpt, “The Limits of AI Understanding”).

Sam Harris

AI and the New Face of Antisemitism (Ep. 453)

Sam Harris speaks with Judea Pearl about causality, AI, and antisemitism. They discuss why LLMs won’t spawn AGI, alignment concerns in the race for AGI, Pearl’s public life after the murder of his son Daniel, the post-October 7th shift toward open anti-Zionism, the overlap between anti-Zionism and antisemitism, the misuse of “Islamophobia,” Israel’s fracture under Netanyahu, confronting anti-Zionism in universities, and other topics…

Listen now

10 months ago · 135 likes · 33 comments · Sam Harris

• The Book of Why, with Dana Mackenzie, 2018.

Ted Sanders

• Allyn-Feuer and Sanders, “Transformative AGI by 2043 is <1% likely,” June 2023. https://arxiv.org/abs/2306.02519

• EA Forum post. https://forum.effectivealtruism.org/posts/ARkbWch5RMsj6xP5p

Events cited in the piece

• OpenAI, “An OpenAI model has disproved a central conjecture in discrete geometry,” May 20, 2026. https://openai.com/index/model-disproves-discrete-geometry-conjecture/

• Alon, Bloom, Gowers, Litt, Sawin, Shankar, Tsimerman, Wang, Wood, “Remarks on the disproof of the unit distance conjecture,” arXiv:2605.20695.

• OpenAI, “Ten advances in mathematics and theoretical computer science,” Aug 1, 2026. https://openai.com/index/ten-advances-in-mathematics/ (Lean certificates: https://github.com/openai/ten-proofs)

• ARC Prize, “OpenAI’s GPT-6 Astra on ARC-AGI-3.” https://arcprize.org/blog/astra

So, last month I observed a self-imposed restraint, a limiting principle that I dubbed “September Smiles” (or occasionally “Smiles Sept”). I outlined it roughly as: no negativity on the notes tab or…

Read more

Read more

Read more

Read more

Read more

Read more

Read more

Read more

I have spoken about Ascension before, but in short: Notables are people of sufficient notoriety, s…

Read more

Read more

Read more