Dan Luu published an audit on September 1, 2026 of roughly 28 dated predictions by Ed Zitron, the most widely cited AI skeptic in circulation, spanning February 2024 to November 2025. Every prediction with a resolvable outcome came back wrong. The post reached 459 points and 536 comments on Hacker News the same day, one of the most-discussed items on the site.
Key facts
- Roughly 28 dated, falsifiable predictions checked; every resolvable one marked wrong.
- The list includes repeated calls that generative AI had permanently peaked (March 2024, July 2024, August 2024, January 2025, April 2025, August 2025, November 2025), that OpenAI's growth was collapsing, and that Cursor would die or sell at a fire-sale price.
- Published September 1, 2026 at danluu.com/zitron; 459 points and 536 comments on Hacker News.
The method is the interesting part, and it is not a rebuttal of Zitron's conclusions so much as an inspection of his arithmetic. Luu takes one prediction apart in detail before listing the rest: a November 2024 talk in which Zitron called Meta "a dying product, and it's kind of a dying company" and argued Google and Microsoft were shoving AI everywhere out of desperation because they no longer knew how to grow. Luu then puts the reported financials next to the claim. Meta went from $135 billion in revenue in 2023 to $201 billion in 2025, with operating profit rising from $47 billion to $83 billion. Alphabet went from $307 billion to $403 billion. Microsoft went from $228 billion to $305 billion. All three grew revenue every year through the period they were supposedly dying.
The sourcing detail matters more than the totals. For Meta's supposed decline, Luu notes, Zitron used monthly-active-user figures from the third-party analytics firm Similarweb rather than Meta's own reported numbers -- and Meta's reported figures show no such sustained decline. This is the pattern Luu argues runs through the work: numbers present, sourced, footnoted, and not connected to the argument they are asked to support. "I suspect he's relying on people's eyes glazing over when they see numbers and just not thinking about what the numbers mean," Luu writes.
Others have found the same thing in narrower spots. Luu quotes Juho Snellman: "if you follow them down to the primary source what they're saying is very different from what Zitron is implying." He cites Timothy B. Lee's examination of a spreadsheet Zitron used to project Anthropic's revenue, which turned out to skip February 1-10, count March 1-10 twice, treat August 21 through October 21 as one month instead of two, and -- per another commenter -- contain a February 30.
The list itself is the payload. "I believe we're reaching the upper limits about what generative AI can do" (February 2024). "Generative AI is a dead-end technology that has peaked" (August 2024). Anthropic reaching $34.5 billion in 2027 revenue is "laughable" (February 2025) -- a claim now sitting awkwardly next to Anthropic's reported run-rate passing $30 billion. Gemini reaching 500 million users is "a number so unrealistic that someone at Google should have been fired" (February 2025); Gemini passed 750 million. "It's pretty easy to come to the conclusion that Cursor is going to die" (July 2025); Cursor exited at $60 billion. Asked in October 2025 when the AI bubble would pop, Zitron answered "no later than Q2 2026."
Luu's framing is not that skepticism is wrong but that this particular skepticism is structurally unfalsifiable in practice. Predicting that progress stops is the mirror image of predicting infinite progress: when it fails you move the date and repeat, and each repetition plays well to an audience that already agrees. He quotes Michał Zalewski on why: "The surest way to build a popular following is to articulate positions that are crisp, strong, and leave no room for doubt... If you take a provocative, edgy stance, you get more attention and likes, so you sort of... self-radicalize?"
Luu also anticipates the accusation and disarms it early: he holds no direct stake in AI companies, does not work at a lab, and published a 2022 audit finding that respected futurists including Ray Kurzweil were generally wrong on both predictions and reasoning. He rates Zitron's reasoning quality as roughly average compared to those futurists -- which reads as a compliment only until you remember every one of them was also wrong.
The honest caveat, and it is a real one: a scorecard restricted to dated, falsifiable claims will systematically favor the auditor. A critic's most valuable contributions are often directional and hard to score -- concerns about circular vendor financing, unsustainable capital expenditure, or the gap between demos and deployed value do not resolve on a date. Luu's own note that he "didn't attempt to catalogue statements that are nonsensical or were simply factually incorrect statements at the time" cuts both ways: it excludes some of Zitron's worst claims, but it also means the selection is the auditor's. And "wrong so far" is not "wrong": a bubble call made in 2024 is not refuted by the bubble not having popped by 2026, only by it never popping. What the audit does establish, and establishes well, is narrower and still damaging -- that the specific numbers used to support these calls frequently do not survive contact with the primary sources they cite.
Originally published on Ground Truth, where every claim is checked against the primary source.
