I would support this Patreon. Particularly to the extent that something like that allows the funding vehicle to remain behind the scenes / separate from the main publication you’ve so lovingly polished and shared with the world over all these years.
4,064 karma · joined January 23, 2022
https://www.popularmechanics.com/culture/gaming/a28954/new-j...
A full-scale arcade version in this genre, evolving since 1996. Realistic controls, some seem even to include train crew uniforms you can wear while driving…
https://www.anthropic.com/news/claude-fable-5-mythos-5
> Today we’re launching Claude Fable 5: a Mythos-class model that we’ve made safe for general use.
It then goes on to a lengthy and detailed section outlining the safety considerations:
https://www.anthropic.com/news/claude-fable-5-mythos-5#:~:te...
https://news.ycombinator.com/item?id=47264756 (4 months ago, 498 points, 392 comments)
Appears to be a forensic walkthrough, working backward to calculate a decryption key to read application logs, working from a disk image of a Windows/IIS machine suspected to be malware-affected.
It includes the output of a run of a utility the author built to help. Which, I mean... I can see why Safe Browsing might get grumpy from the content alone; I'm also in no position to assess whether there's actual sneaky stuff going on too.
Excerpt:
Published: 23/01/2026
I recently had someone reach out to me with an interesting problem. They had found a 1316 event in their Windows application logs that contained a likely malicious view state. There was just one catch, it was encrypted. [...] They were able to dump the autogen keys from the Windows registry, however they didn’t know how to use these to decrypt their view state.
[...]
In this post, I’ll be covering:
* How the autogen keys are generated
* How the master machine keys are derived from the autogen keys
* How the final machine keys are derived from the master machine keys
* How the final keys can be used to decrypt view state messages
The specific image that comes to mind, though—I remember dozens upon dozens of ambulances lined up at the ready along the waterfront, their crews hanging out enjoying the madness until they would be called upon.
Having been raised in pretty cautious circles, that image struck me: the powers-that-be knew some proportion of bad stuff would happen. I figured that meant they’d “just make a law against it.” Instead of trying to stop it, though, they focused on preparedness and timely response.
Since then, of course, I’ve learned that The Law has a lot less raw power than I’d imagined against tradition and popular will…
Since then I’ve accepted as creed that the best and most human parts of life reach their most salient in the moments of paradox between principles. “Avoid safety risks” is correct, “living fully requires accepting risks” is also correct.
[0] https://www.barcelona-tourist-guide.com/en/events/sant-joan/...
https://www.nytimes.com/2026/06/20/business/egg-prices-down....
And Levine’s column, which Stoller links to (with rather less color commentary):
https://www.bloomberg.com/opinion/newsletters/2026-07-01/egg...
It feels like Mr. Stoller spends a lot of time here insinuating that because price manipulation happened on the margins of this supply crunch, there was no supply crunch, and everything’s just moustache-twirling tycoon conspiracies:
> While most normal people at the time thought someone was likely scamming them, that is not the message you heard from the industry, elite media, or economists.
> In 2025, egg prices dropped dramatically […] these price declines suggested that supply and demand were doing their magical work. Populists were mocked as ignoring natural market forces. […but…] It turns out, when [these conspirators] felt threatened by legal action, the alleged price-fixing stopped. Suddenly, the avian flu epidemic was no longer pushing up prices.
I mean… it can very much be both. Slaughtering all the chickens really can reduce the number of eggs in the world, people really can be willing to pay more for the few that are left, you really will get more eggs again when you make enough new chickens and wait til they grow to egg-pooping age. Even as it was also true that some greedy people’s unfair play magnified the dynamics that were already happening.
But like—even at the higher prices, eggs weren’t going unsold at the end of the day.
To me this whole thing still feels like things working the way the dastardly elite theorists suggest it does: the reason we treat collusion like this as bad and illegal in the first place—besides the casual sense of grossness and unfair play—is that the misleading signal provokes overproduction and therefore a price collapse.
The price did indeed go on to collapse by 93% to pennies a dozen; that’s squeezing farmers brutally.
The investigators investigated, the prosecutors prosecuted, the manipulative behavior stopped, the contracts got adjusted, the price index mechanism got revisited…
I feel like the error is similar to what bothers me listening to day-trader types: conflating raw synthetic-price-index movements with the underlying physical reality they represent.
> We hope that Apple will take steps to limit the attack surface area even before the vulnerability is fixed. Disabling creation of new Hide My Email addresses could be helpful. It also seems responsible to notify all Hide My Email users of the risk.
Thank you for your work, and your persistence against our Sphinx-like overlords!
What’s described sounds like it might be spooky. It might also be a magic trick to some degree… Mr. Cox’s PoC—“I gave a fresh Hide-My-Email alias to a guy who knows who I am, and he told me the email on my Apple ID”—is consistent with the claimed behavior but not exactly watertight.
It also sounds like it might be the sort of thing that’s either “just how the email ecosystem works” or mitigable by covert means. For example, if Apple can identify exploit attempts from its privileged vantage over its infrastructure, maybe that’s the basis for its relaxed impact assessment.
I’m reminded of Amazon’s risk assessment with respect to some Quick bug recently [0]: “yeah, it’s bad, but we checked and there are literally zero people other than you who’ve ever used that feature that way.”
Or maybe it’s the kind of thing that requires a structural sort of tradeoff to conclusively fix. I could imagine the exposure mechanism having something to do with their forthcoming move to segregate aliases to their own “private.icloud.com” domain.
(A move at which Mr. Cox swipes in the 404 Media article, too, of course, but hey—“impact journalism.”)
And then, since we have only vibes to go on, there’s the judgment reflected in the researcher’s email to Apple:
> “It seems that ending new sales of Hide My Email until the problem is fixed would be an effective way to limit the number of customers at risk. Is that an option?” Murphy wrote back.
I can only hope that was a sardonic moment of frustration quoted out of context… Hide My Email is “sold” as a tiny tiny bonus feature of a much bigger iCloud+ product. But as-quoted, it’s giving a little bit of Chicken Little… I’m reminded of the time somebody demanded that a firm I’m familiar with halt all sales (and pay hush money) because of a CRITICAL SECURITY HOLE: you could access the contents of a password field by typing the password in the field, pressing F12 in the browser, and typing $(“#pw-input”).value …
If the flaw really is the sort of thing that required fundamental product changes to fully address—like this domain segregation thing—a year doesn’t seem wild at all to make that transition safely and at scale. Especially if they identified effective mitigations in the meantime.
Then again, maybe they really are negligent…
[0] https://www.theregister.com/columnists/2026/05/13/aws-patche...
It would be fine if you gave me a beautiful and whimsical way to choose to express my feelings, and I took it. But when you're disguising the flag in code as an "EXTRATREE," that signals to me that you're trying to slip through a surprise without my noticing:
#ifdef USE_EXTRATREE
if (global.Language && !strcmp(global.Language,"ru") && drand48() < 0.3)
tt = MAXTREETYPE;
if (drand48() < 0.02)
tt = MAXTREETYPE;
#endif
I'm reminded of a situation I encountered some years ago where a person opened a web browser in front of a classroom—no porn in their history, nothing untoward, just going to a high-profile mainstream news site or something in service of a classroom discussion—and all the targeted ads were for things like HIV medicines and mainstream campaigns choosing ad variants that depicted gay couples.
Not the time or the place, that person didn't ask for it, and it led to deep consequences for them—from "outing" on one side, to accusations of "grooming that classroom full of students" and "probably being riddled with AIDS" on the other—that this good, responsible, kind, wise person did nothing to invite.
The targeters probably thought they were doing something righteous or even "accepting" by "making sexual minorities feel seen" or something—but by putting words in the person's mouth without their consent or agency, they caused great unnecessary harm to somebody who didn't deserve it.
As in, as long as I clean up really well afterward, I can pretty much do what I want?
I imagine reasons for what you’re asking might include:
* Prompting an LLM is work, and they’re already overworked just doctoring—every conversation with a computer is a conversation you’re not having with a patient;
* They’re probably right more often than they’re wrong;
* “When you hear hooves, think horses, not zebras”: the 15th case today of strep throat is probably strep throat, regardless of today’s 15th falsely-confident LLM weighing-up;
* They tend to have spent many many years honing a clinical intuition that makes an examination, to some degree, hard to articulate fully to the LLM;
* Liability/overdiagnosis: All this stuff is probabilistic. Inevitably, there’s going to be a time when the LLM throws out something I thought unlikely that turns out to be right, and there will be other times when it’s wrong but now I have to document why. How many false leads do I need to chase per one true differential? Does this really compare favorably to seeking a second opinion from another human doctor?
* Not everything needs to make it into the record. Once it’s in the LLM, it’s discoverable and litigable and hackable and permanent;
* Medicine is practiced in very different ways in different contexts—even in this thread, one radiologist routinely orders ultrasounds for soft tissue shoulder problems, and the other medical-world person replying has never heard of such a thing—presumably both within US health care contexts. Some doctors hand out antibiotics like candy, others are more cautious with respect to resistance. What’s right can depend on the time, the place, the clinical setting—more than just the immediate patient-level facts at hand, in ways that become awkward or unwise to express explicitly.
And of course… who’s to say they don’t do LLM-assisted research, in cases where they think it might be helpful?
Which is not to say let’s not do it anyway and see!
See also contemporaneous reaction at:
https://news.ycombinator.com/item?id=48624782 (6 days ago, 244 points, 133 comments)
I seem to recall legal commentators reacting with an eyeroll—apparently judges split much finer hairs than these for a living—but it was a cute stunt.
[1] https://m.youtube.com/watch?v=sJtm0MoOgiU and https://www.the-independent.com/tech/music-copyright-algorit...
Maybe the chatbot can reproduce variations on technical artifacts with similar formal characteristics as the Candy Crush game. But the reproductions don’t come with the essence of what made it important: its cultural novelty, the whimsy of its graphics, the freshness of its addictive mechanism, its arrival when people had smartphones and were fishing around for what to do with them, the fact that everybody else was discovering it along with you (and talking about that fact), its timing before its competitors dialed in even more potent attention sinks…
“Voici mon secret. Il est très simple: On ne voit bien qu'avec le cœur. L'essentiel est invisible pour les yeux.”
I can make a photo or video with my phone camera anywhere, anytime, and I can do so essentially for free, in essentially unlimited quantity. Then Sean Baker turns around and makes Tangerine with his iPhone 5S… and if anything, I’m even more sensitized to the way good photography has something above and beyond what I’m doing.
Art in the Age of Mechanical Reproduction [0] comes to mind.
This would be shortly after Wynn-Williams’ 2017 departure from the policy role she describes in the book at hand. And it would be around the time that word was getting around about Facebook’s role in what Amnesty and the UN described as a campaign of ethnic cleansing against the Rohingya minority in Myanmar. Among other things.
That story didn’t go on to a happy ending after 2017, and one imagines that, in the decade since, there have been fewer and fewer situations of strife, geopolitical gamesmanship, and civil conflict where Facebook/Instagram/Whatsapp could avoid taking consequential policy decisions.
Meta were enmeshed in ample US domestic drama during that period, too; and Clegg’s replacement in the policy role was someone in the new administration’s orbit. Perhaps that’s fresh on the ancestor commenters’ minds.
Given the culture Ms. Wynn-Williams describes—and the tumult of the decade when he held the policy role—one imagines Mr. Clegg might have some stories to share should he choose to…
Assuming, of course, said state agency is operating under sufficiently strategic governance and management…
https://www.gartner.com/en/research/methodologies/gartner-hy...
The idea is there’s a rush of irrational exuberance when an “innovation trigger” makes a new toy looks promising, and everybody rushes to use it for everything, regardless of whether its suitability-for-purpose is proven. Inevitably many of those pioneers find that it’s not good for their particular problems after all; usage reaches a “peak of inflated expectations,” and crashes into a “trough of disillusionment.”
Then the tech enters a quieter and more gradual “slope of enlightenment” as people work out use cases where the tech actually adds value; then adoption reaches a “plateau of productivity.”
Worth a glance at the way they map this to prior waves of technological exuberance.
The Biggest And Richest have Mythos. But does that diminish the utility, to me personally, of my open-weights model? Or even my humble-individual-grade Opus subscription? Relative to what I could do three years ago, computationally speaking?
Does a pretty good life have to involve controlling more material relative to members of a specific class, rather than just… deciding what’s enough, materially speaking; and stopping there?
E.g., https://en.wikipedia.org/wiki/Stop_and_identify_statutes ?
Until, of course, they learn to like the replacement oil. At which point, break back out the DEET!
Another is the idea that schools are motivated by money in the same way profit-seeking ventures are. A company’s shareholders might respond to financial threats and incentives, but the teachers on district-regulated wages? What’s the phrase, can’t squeeze blood from a turnip?
Then there’s of course the construct validity of standardized tests as a measure of “suckiness”—they’re easy to administer at scale and to compare across years and between schools—but do they really capture every flavor of good work that’s done at a school? They’re the best thing we have, but does that make them good enough?
The main issue, though, I think we can frame in terms of a slightly different legibility issue: since the school is the only variable we directly control, we model the school’s “suckiness” as a function of its… what, budget? Staff bonuses? Whoever exactly is it who we’re proposing to punish by removing funds? But just as I imagine we can think of kids who would be fine either way—one of the less provocative stereotypes that comes to mind is that of a Tiger Mom kind of community—we can probably think of kids who won’t be fine. The less provocative stereotype that comes to mind is a child with special needs: with an aide, maybe that child may develop enough to participate in society, and we’re a more humane and moral society for trying. For that matter there are other children who are living and growing up in situations where survival is always going to come before their test scores—and those are probably the students with guardians least equipped to exercise “school choice.” How does punishing their school improve those kids’ outcomes?
Often students who perform poorly need more resources, not fewer.
…are a few of the counterarguments, anyway.
I feel like this trope is strongest amongst musicians-feeling-underappreciated, but that the idea seeps in to all manners of creative work: that, because you’re rightfully proud of what you do, the audience is wronging you (or “lazy,” or “sad,” or “cheap,” or “tasteless”) by not appreciating it. It doesn’t make me feel a lot of sympathy.
For a taste, Joanna Steinhardt for the LA Review:
https://lareviewofbooks.org/article/the-mycophiles-plea-on-m...
I suspect some folks here might also appreciate his early-career (2011) musical collaboration with brother Cosmo Sheldrake and friends, as the Gentle Mystics: