A “reasoning AI” is a large language model — a chatbot — that gives you a list of steps it took to reach a conclusion. Or a list of steps it says it took.

LLMs don’t know what a fact is — they just generate words. So chatbots are notorious for hallucinating — when it generates a pile of plausible text that’s full of factual errors, but it’s the right shape for whatever you were asking for.

Given this, it’s no surprise that when you ask a “reasoning” model to explain its reasoning steps, it just generates something that looks like a list of reasoning steps — and not necessarily anything it actually used.

Anthropic is an LLM vendor founded by AI doom mongers who left OpenAI. They’re extremely into the idea that the default behaviour of AI is that the robot is plotting against us. Anthropic was responsible for a pile of preprints claiming an LLM was trying to deceive researchers, and in every case it turned out the researchers told the LLM to behave that way.

Even Anthropic couldn’t really spin this present study — of course the BSing chatbot was going to BS about how it was BSing! Though their paper talks about “faithfulness” instead of, say, “accuracy,” because they just can’t stop anthropomorphising the chatbot. [Anthropic; Anthropic, PDF]

Other researchers looking at reasoning LLM have found the same behaviour — that everything an LLM puts out is confabulated, so when it explains its behaviour, that’s confabulated too. Because they’re confabulation machines, not true-or-false machines.

Transluce is a San Francisco “AI safety” charity which brags that it analyses LLMs using other LLMs. Transluce was shocked to find that OpenAI’s o3 model “frequently fabricates actions it took to fulfill user requests, and elaborately justifies the fabrications when confronted by the user.” [Transluce; Transluce]

Transluce doesn’t have a visible cohort of AI doomers, but they do insist on describing this behaviour, not as inaccuracy, but as a not being “truthful” — even as an LLM has no concept of true or false. It doesn’t even have concepts. [Twitter, proxy]

Benj Edwards at Ars Technica wrote up Anthropic’s findings. The first version of his story cut’n’pasted Anthropic’s humanised descriptions of the chatbot so egregiously that an editor fixed the headline and thoroughly revised the story to “reduce overly anthropomorphic language.” [Ars Technica, archive of 10 April 2025]

This is a good worked example of gullible AI journalism and how to fix it a little. Here are my three favourite changes. First example:

• Old version: “they sometimes hide their actual methods while fabricating elaborate explanations instead.”
• New version: “the ‘work’ they show can sometimes be misleading or disconnected from the actual process used to reach the answer.”

Second example:

• Old version: “OpenAI’s o1 and o3 series SR models deliberately obscure the accuracy of their ‘thought’ process, so this study does not apply to them.”
• New version: these “SR models were excluded from this study.”

For the third example, Edwards went so over-the-top that the editor had to insert a whole new paragraph about how “It’s important to note that AI models don’t have intentions or desires”.

Let’s hope Ars does better going forward.

• Video version

Today's links

• Mark Zuckerberg personally lost the Facebook antitrust case: A folder on his desktop called "mens rea."
• Hey look at this: Delights to delectate.
• Object permanence: 2005, 2010, 2015, 2020, 2024
• Upcoming appearances: Where to find me.
• Recent appearances: Where I've been.
• Latest books: You keep readin' em, I'll keep writin' 'em.
• Upcoming books: Like I said, I'll keep writin' 'em.
• Colophon: All the rest.

Mark Zuckerberg personally lost the Facebook antitrust case (permalink)

It's damned hard to prove an antitrust case: so often, the prosecution has to prove that the company intended to crush competition, and/or that they raised prices or reduced quality because they knew they didn't have to fear competitors.

It's a lot easier to prove what a corporation did than it is to prove why they did it. What am I, a mind-reader? But imagine for a second that the corporation in the dock is a global multinational. Now, imagine that the majority of the voting shares in that company are held by one man, who has served as the company's CEO since the day he founded it, personally calling every important shot in the company's history.

Now imagine that this founder/CEO, this accused monopolist, was an incorrigible blabbermouth, who communicated with his underlings almost exclusively in writing, and thus did he commit to immortal digital storage a stream – a torrent – of memos in which he explicitly confessed his guilt.

Ladies and gentlepersons, I give you Mark Zuckerberg, founder and CEO of Meta (nee Facebook), an accused monopolist who cannot keep his big dumb fucking mouth shut.

At long, long last, the FTC's antitrust trial against Meta is underway, and this week, Zuck himself took the stand, in agonizing sessions during which FTC lawyers brandished printouts of Zuck's own words before him, asking him to explain away his naked confessions of guilt. It did not go well for Zuck.

In a breakdown of the case for The American Prospect, editor-in-chief David Dayen opines that "The Government Has Already Won the Meta Case," having hanged Zuck on his own words:

https://prospect.org/power/2025-04-16-government-already-won-meta-case-tiktok-ftc-zuckerberg/

The government is attempting to prove that Zuck bought Instagram and Whatsapp in order to extinguish competitors (and not, for example, because he thought they were good businesses that complemented Facebook's core product offerings).

This case starts by proving how Zuck felt about Insta and WA before the acquisitions. On Insta, Zuck circulated memos warning about Insta's growth trajectory:

they appear to be reaching critical mass as a place you go to share photos

and how that could turn them into a future competitor:

[Instagram could] copy what we’re doing now … I view this as a big strategic risk for us if we don’t completely own the photos space.

These are not the words of a CEO who thinks another company is making a business that complements his own – they're confessions that he is worried that they will compete with Facebook. Facebook tried to clone Insta (Remember Facebook Camera? Don't feel bad – neither does anyone else). When that failed, Zuck emailed Facebook execs, writing:

[Instagram's growth is] really scary and why we might want to consider paying a lot of money for this.

At this point, Zuck's CFO – one of the adults in the room, attempting to keep the boy king from tripping over his own dick – wrote to Zuck warning him that it was illegal to buy Insta in order to "neutralize a potential competitor."

Zuck replied that he was, indeed, solely contemplating buying Insta in order to neutralize a potential competitor. It's like this guy kept picking up his dictaphone, hitting "record," and barking, "Hey Bob, I am in receipt of your memo of the 25th, regarding the potential killing of Fred. You raise some interesting points, but I wanted to reiterate that this killing is to be a murder, and it must be as premeditated as possible. Yours very truly, Zuck."

Did Zuck buy Insta to neutralize a competitor? Sure seems like it! For one thing, Zuck cancelled all work on Facebook Camera "since we're acquiring Instagram."

But what about after the purchase. Did Zuck reduce quality and/or raise costs? Well, according to the company, it enacted an "explicit policy of not prioritizing Instagram’s growth" (a tactic called "buy or bury"). At this juncture, Zuckerberg once again put fingers to keyboard in order to create an immortal record of his intentions:

By not killing their products we prevent everyone from hating us and we make sure we don’t immediately create a hole in the market for someone else to fill.

And if someone did enter the market with a cool new gimmick (like, say, Snapchat with its disappearing messages)?

Even if some new competitors spring up, if we incorporate the social mechanics they were using, these new products won’t get much traction since we’ll already have their mechanics deployed at scale.

Remember, the Insta acquisition is only illegal if Zuck bought them to prevent competition in the marketplace (rather than, say, to make a better product). It's hard to prove why a company does anything, unless its CEO, founder, and holder of the majority of its voting stock explicitly states that his strategy is to create a system to ensure that innovating new products "won't get much traction" because he'll be able to quickly copy them.

So we have Zuck starving Insta of development except when he needs to neutralize a competitor, which is just another way of saying he set out to reduce the quality of the product after acquisition, a thing that is statutorily prohibited, but hard to prove (again, unless you confess to it in writing, herp derp).

But what about prices? Well, obviously, Insta doesn't charge its end-users in cash, but they do charge in attention. If you want to see the things you've explicitly asked for – posts from accounts you follow – you have to tolerate a certain amount of "boosted content" and ads, that is, stuff that Facebook's business customers will pay to nonconsensually cram into your eyeballs.

Did that price go up? Any Insta user knows the answer: hell yes. Instagram is such a cesspit of boosted content and ads that it's almost impossible to find stuff you actually asked to see. Indeed, when a couple of teenagers hacked together an alternative Insta client called OG App that only showed you posts from accounts you followed, it was instantly the most popular app on Google Play and Apple's App Store (and then Google and Apple killed it, at Meta's request):

https://pluralistic.net/2023/02/05/battery-vampire/#drained

But why did the price go up? Did it go up because Facebook had neutralized a competitor by purchasing it, and thus felt that it could raise prices without losing customers? Again, a hard thing to prove…unless Zuck happened to put it in writing. Which he did, as Brendan Benedict explains in Big Tech On Trial:

I think we’re badly mismanaging this right now. There’s absolutely no reason why IG ad load should be lower than FB at a time when . . . we’re having engagement issues in FB. If we were managing our company correctly, then at a minimum we’d immediately balance IG and FB ad load . . . But it’s possible we should even have a higher ad load on IG while we have this challenge so we can replace some ads with [People You May Know] on FB to turn around the issues we’re seeing.

https://www.bigtechontrial.com/p/zuckerberg-v-zuckerberg-will-the

So there you have it: Zuck bought Insta to neutralize a competitor, and after he did, he lowered its quality and raised its prices, because he knew that he was operating without significant competitors thanks to his acquisition of that key competitor. Zuck's motivations – as explained by Zuck himself – were in direct contravention of antitrust law, a thing he knew (because his execs explained it to him). That's a pretty good case.

But what about Whatsapp? How did Zuck feel about it? Well, he told his board that Whatsapp was Facebook's greatest "consumer risk," fretting that "Messenger isn’t beating WhatsApp." He blocked Whatsapp ads on Facebook, telling his team that it was "trying to build social networks and replace us." Sure, they'd lose money by turning away that business, but the "revenue is immaterial to us compared to any risk." Sure seems like Zuck saw Whatsapp as a competitor.

Meta's final line of defense in this case is that even if they did some crummy, illegal things, they still didn't manage to put together a monopoly. According to Meta's lawyers – who're billing the company more than $1m/day! – Meta is a tiny fish in a vast ocean that has many competitors, like Tiktok:

https://www.levernews.com/mr-zuckerbergs-very-expensive-day-in-court/

There's only one problem with this "market definition" argument, and that problem's name is Chatty Mark Zuckerberg. On the question of market definition, FTC lawyers once again raised Zuckerberg's own statements and those of his top lieutenants to show that Zuckerberg viewed his companies as "Personal Social Networks" (PSNs) and not as just generic sites full of stuff, competing with Youtube, Tiktok, and everyone else who lets users post things to the internet.

Take Instagram boss Adam Mosseri, who explained that:

Instagram will always need to focus on friends and can never exclusively be for public figures or will cease to be a social product.

And then there was Zuck's memo explaining why he offered $6b for Snapchat:

Snap Stories serves the exact same use case of sharing and consuming feeds of content that News Feed and Instagram deliver. We need to take this new dynamic seriously—both as a competitive risk and as a product opportunity to add functionality that many people clearly love and want to use daily.

And an internal strategy document that explained the competitive risks to Facebook:

Social networks have two stable equilibria: either everyone uses them, or no-one uses them. In contrast, nonsocial apps (e.g. weather apps, exercise apps) can exist [somewhere] along a continuum of adoption. The binary nature of social networks implies that there should exist a tipping point, ie some critical mass of adoption, above which a network will organically grow, and below which it will shrink.

Sure sounds like Facebook sees itself as a "social network," and not a "nonsocial app." And of course – as Dayen points out – when Tiktok (a company Meta claims as a competitor) went up for sale, Meta did not enter a bid, despite being awash in free cash flow.

In Zuckerberg's defense, he's not the only tech CEO who confesses his guilt in writing (recall that FTX planned its crimes in a groupchat called WIREFRAUD). Partly that's because these firms are run by arrogant twits, but partly it's because digital culture is a written culture, where big, dispersed teams expected to work long hours from offices all over the world as well as from their phones every hour of day and night have to rely on memos to coordinate:

https://pluralistic.net/2023/09/03/big-tech-cant-stop-telling-on-itself/

When Dayen claims that "the government has won the Meta case," he doesn't mean the judge will rule in the FTC's favor (though there's a high likelihood that this will happen). Rather, he means that the case has been proven beyond any kind of reasonable doubt, in public, in a way that has historically caused other monopolists to lose their nerve, even if they won their cases. Take Microsoft and IBM – though both companies managed to draw out their cases until a new Republican administration (Reagan for IBM, GWB for Microsoft) took office and let them off the hook, both companies were profoundly transformed by the process.

IBM created the market for a generic, multivendor PC whose OS came from outside the company:

https://www.eff.org/deeplinks/2019/08/ibm-pc-compatible-how-adversarial-interoperability-saved-pcs-monopolization

And Microsoft spared Google the same treatment it had meted out to Netscape, allowing the company to grow and thrive:

https://apnews.com/article/google-apple-microsoft-antitrust-technology-cases-1e0c510088825745a6e74ba3b81b44c6

Trump being Trump, it's not inconceivable that he will attempt to intervene to get the judge to exonerate Meta. After all, Zuck did pay him a $1m bribe and then beg him to do just that:

https://gizmodo.com/zuckerberg-really-thought-trump-would-make-metas-legal-problems-go-away-2000589897

But as Dayen writes, the ire against Meta's monopolistic conduct is thoroughly bipartisan, and if Trump was being strategic here (a very, very big "if"), he would keep his powder dry here. After all, if the judge doesn't convict Meta, Trump won't have wasted any political capital. And if Meta is convicted, Trump could solicit more bribes and favors at the "remedy" stage, when a court will decide how to punish Meta, which could be anything from a fine to a breakup order, to a nothingburger of vague orders to clean up its act.

Hey look at this (permalink)

• Unions, Not Just Factories, Will Make America Great https://www.laborpolitics.com/p/unions-not-just-factories-will-make

• How Wales is building a sharing economy through its ‘libraries of things’ https://theprogressplaybook.com/2025/04/05/how-wales-is-building-a-sharing-economy-through-its-libraries-of-things/ (h/t Bill McKibben)

• Google Is Operating an Ad Tech Monopoly, Judge Rules https://gizmodo.com/google-is-operating-an-ad-tech-monopoly-judge-rules-2000590754

Object permanence (permalink)

#20yrsago White House drug agency giving away anti-drug ringtones https://web.archive.org/web/20050419024323/http://www.freevibe.com/stepup/freevibe_ringtones.asp

#15yrsago George Washington owes $100K in library fines https://www.loweringthebar.net/2010/04/president-accused-of-theft-failing-to-pay-massive-library-fines.html

#10yrsago Slack’s CEO gives charmingly bullshit-free justification for $2.8B valuation https://archive.nytimes.com/bits.blogs.nytimes.com/2015/04/16/is-slack-really-worth-2-8-billion-a-conversation-with-stewart-butterfield/

#5yrsago Delivery services are gouging restaurants to death https://pluralistic.net/2020/04/18/politics-of-discouragement/#rent-seekers

#5yrsago ICANN pauses selloff of .ORG registry https://pluralistic.net/2020/04/18/politics-of-discouragement/#savedotorg

#5yrsago Garbage Math https://pluralistic.net/2020/04/18/politics-of-discouragement/#xkcd-2295

#5yrsago The Rent and Mortgage Cancellation Act https://pluralistic.net/2020/04/18/politics-of-discouragement/#ilhan-omar

#1yrago Podcasting "Capitalists Hate Capitalism" https://pluralistic.net/2024/04/18/in-extremis-veritas/#the-winnah

Upcoming appearances (permalink)

• Auckland: Unity Books, May 2, 6PM
  https://www.eventbrite.co.nz/e/an-evening-with-cory-doctorow-tickets-1320740102199

• Wellingon: Unity Books, May 3, 3PM
  https://www.unitybooks.co.nz/news-and-events/author-talk-picks-and-shovels-by-cory-doctorow

• Pittsburgh: Picks and Shovels at White Whale Books, May 15
  https://whitewhalebookstore.com/events/20250515

• Pittsburgh: PyCon, May 16
  https://us.pycon.org/2025/schedule/

• PDX: Teardown 2025, Jun 20-22
  https://www.crowdsupply.com/teardown/portland-2025

• PDX: Picks and Shovels with bunnie Huang at Barnes and Noble, Jun 20
  https://stores.barnesandnoble.com/event/9780062183697-0

• London: How To Academy with Riley Quinn, Jul 1
  https://howtoacademy.com/events/cory-doctorow-the-fight-against-the-big-tech-oligarchy/

• Manchester: Picks and Shovels at Blackwell's Bookshop, Jul 2
  https://www.eventbrite.co.uk/e/an-evening-with-cory-doctorow-tickets-1308451968059

• Manchester: Co-operatives UK Co-op Congress keynote, Jul 3
  https://www.uk.coop/events-and-training/events-calendar/co-op-congress-2025-book-your-place

• New Orleans: DeepSouthCon63, Oct 10-12, 2025
  http://www.contraflowscifi.org/

Recent appearances (permalink)

• Can we use the Internet for Democracy?
  https://www.youtube.com/watch?v=Zh_HON6iql8

• Fightback Against Trump's Tariff Attack (Avi Lewis)
  https://www.youtube.com/watch?v=P9sgIAc6z_o

• The Voice of Canadian Humanism
  https://open.spotify.com/episode/7uuwdZTIbWzKhBQ3mmMiRv?context=spotify%3Ashow%3A6N5hl8on16CfaeYArrKyqZ

Latest books (permalink)

• • Picks and Shovels: a sequel to "Red Team Blues," about the heroic era of the PC, Tor Books (US), Head of Zeus (UK), February 2025 (https://us.macmillan.com/books/9781250865908/picksandshovels).
• The Bezzle: a sequel to "Red Team Blues," about prison-tech and other grifts, Tor Books (US), Head of Zeus (UK), February 2024 (the-bezzle.org). Signed, personalized copies at Dark Delicacies (https://www.darkdel.com/store/p3062/Available_Feb_20th%3A_The_Bezzle_HB.html#/).

• "The Lost Cause:" a solarpunk novel of hope in the climate emergency, Tor Books (US), Head of Zeus (UK), November 2023 (http://lost-cause.org). Signed, personalized copies at Dark Delicacies (https://www.darkdel.com/store/p3007/Pre-Order_Signed_Copies%3A_The_Lost_Cause_HB.html#/)

• "The Internet Con": A nonfiction book about interoperability and Big Tech (Verso) September 2023 (http://seizethemeansofcomputation.org). Signed copies at Book Soup (https://www.booksoup.com/book/9781804291245).

• "Red Team Blues": "A grabby, compulsive thriller that will leave you knowing more about how the world works than you did before." Tor Books http://redteamblues.com. Signed copies at Dark Delicacies (US): and Forbidden Planet (UK): https://forbiddenplanet.com/385004-red-team-blues-signed-edition-hardcover/.

• "Chokepoint Capitalism: How to Beat Big Tech, Tame Big Content, and Get Artists Paid, with Rebecca Giblin", on how to unrig the markets for creative labor, Beacon Press/Scribe 2022 https://chokepointcapitalism.com

• "Attack Surface": The third Little Brother novel, a standalone technothriller for adults. The Washington Post called it "a political cyberthriller, vigorous, bold and savvy about the limits of revolution and resistance." Order signed, personalized copies from Dark Delicacies https://www.darkdel.com/store/p1840/Available_Now%3A_Attack_Surface.html

• "How to Destroy Surveillance Capitalism": an anti-monopoly pamphlet analyzing the true harms of surveillance capitalism and proposing a solution. https://onezero.medium.com/how-to-destroy-surveillance-capitalism-8135e6744d59?sk=f6cd10e54e20a07d4c6d0f3ac011af6b) (signed copies: https://www.darkdel.com/store/p2024/Available_Now%3A__How_to_Destroy_Surveillance_Capitalism.html)

• "Little Brother/Homeland": A reissue omnibus edition with a new introduction by Edward Snowden: https://us.macmillan.com/books/9781250774583; personalized/signed copies here: https://www.darkdel.com/store/p1750/July%3A__Little_Brother_%26_Homeland.html

• "Poesy the Monster Slayer" a picture book about monsters, bedtime, gender, and kicking ass. Order here: https://us.macmillan.com/books/9781626723627. Get a personalized, signed copy here: https://www.darkdel.com/store/p2682/Corey_Doctorow%3A_Poesy_the_Monster_Slayer_HB.html#/.

Upcoming books (permalink)

• Enshittification: Why Everything Suddenly Got Worse and What to Do About It, Farrar, Straus, Giroux, October 7 2025
  https://us.macmillan.com/books/9780374619329/enshittification/

• Unauthorized Bread: a middle-grades graphic novel adapted from my novella about refugees, toasters and DRM, FirstSecond, 2026

• Enshittification, Why Everything Suddenly Got Worse and What to Do About It (the graphic novel), Firstsecond, 2026

• The Memex Method, Farrar, Straus, Giroux, 2026

Colophon (permalink)

Today's top sources:

Currently writing:

• Enshittification: a nonfiction book about platform decay for Farrar, Straus, Giroux. Status: second pass edit underway (readaloud)

• A Little Brother short story about DIY insulin PLANNING

• Picks and Shovels, a Martin Hench noir thriller about the heroic era of the PC. FORTHCOMING TOR BOOKS FEB 2025

Latest podcast: Nimby and the D-Hoppers CONCLUSION https://craphound.com/stories/2025/04/13/nimby-and-the-d-hoppers-conclusion/

This work – excluding any serialized fiction – is licensed under a Creative Commons Attribution 4.0 license. That means you can use it any way you like, including commercially, provided that you attribute it to me, Cory Doctorow, and include a link to pluralistic.net.

https://creativecommons.org/licenses/by/4.0/

Quotations and images are not included in this license; they are included either under a limitation or exception to copyright, or on the basis of a separate license. Please exercise caution.

How to get Pluralistic:

Blog (no ads, tracking, or data-collection):

Pluralistic.net

Newsletter (no ads, tracking, or data-collection):

https://pluralistic.net/plura-list

Mastodon (no ads, tracking, or data-collection):

https://mamot.fr/@pluralistic

Medium (no ads, paywalled):

https://doctorow.medium.com/

Twitter (mass-scale, unrestricted, third-party surveillance and advertising):

https://twitter.com/doctorow

Tumblr (mass-scale, unrestricted, third-party surveillance and advertising):

https://mostlysignssomeportents.tumblr.com/tagged/pluralistic

"When life gives you SARS, you make sarsaparilla" -Joey "Accordion Guy" DeVilla

ISSN: 3066-764X

Michael Kalus posted a photo: