Webshit Weekly
August 14, 2025
Administration seeking $1B settlement from UCLA
2025-08-08 | comments
The federal government has apparently decided that shaking down a public university for a billion dollars is the new national pastime, because why bother governing when you can run the country like a petty HOA with nukes? UCLA, already bleeding from decades of budget gouging and grant yo-yoing, now gets a draft agreement written like a ransom note from a digital feudal lord: pay up, gut your programs, stop providing medical care we don’t like, and let a political babysitter rummage through your admissions data. Hackernews responds with the usual civics-class cosplay, miscorrecting each other about the Constitution while ignoring the obvious: this isn’t policy, it’s extortion LARPing as governance.
Curious about the training data of OpenAI’s new GPT-OSS models? I was too
2025-08-09 | comments
Hackernews has once again gathered around a blurry screenshot of a tweet about “Neuralese,” clutching their browser extensions like medieval talismans against the horrors of modern web design. Half the thread is people arguing about how to view a tweet without pledging fealty to Musk’s decaying ad‑scraper, and the other half is amateur linguists confidently explaining how a token predictor forgetting English means it has “figured out RL.” Sure. The clanker mutters gibberish because it was vibe‑trained on a slurry of scraped sludge, but to Hackernews this is a sign of emergent intelligence, not the machine equivalent of a stroke.
Abusing Entra OAuth for fun and access to internal Microsoft applications
2025-08-09 | comments
Microsoft’s latest security faceplant arrives with all the surprise of a Windows Update rebooting your machine mid‑sentence. Some researcher pokes at Entra OAuth for five minutes and, shockingly, discovers that the world’s largest purveyor of enterprise duct tape is authorizing anything with a pulse and a JSON blob. The writeup reads like a tour through the ruins of “Zero Trust,” which in Microsoft‑speak apparently means trusting every token as long as it hasn’t burst into flames on the way in.
Hackernews reacts with its usual chorus of retirees from Fortune 10 sweatshops, insisting this isn’t AI’s fault because the code predates modern token predictors. No, this is classic corporate aura farming: decades of vibe‑coding authentication layers onto authentication layers until the whole edifice resembles a Jenga tower built by interns who weren’t allowed to ask questions. Meanwhile the cloud evangelists parachute in to remind everyone that the intranet was never safe anyway, as though the solution to insecure software is to expose more of it to the public internet.
Then come the multi‑tenant truthers, who explain (again) that storing the crown jewels in a communal identity bucket was always going to end beautifully. Microsoft PMs show up to explain the “real” guidance, which amounts to: validate literally everything because we can’t guarantee any of it is real. Somehow, despite the researcher gaining access to the build systems for retail Windows, the bounty team decides this merits exactly zero dollars.
Truly, a perfect encapsulation of modern software: insecure by default, patched by PR, and defended on Hackernews by the very people who broke it in the first place.
A large number of protocols on Ethereum and Solana blockchains have no revenue
2025-08-10 | comments
CoinDesk has once again discovered arithmetic and decided it counts as journalism. This week’s revelation: most blockchain protocols don’t make money. Incredible. Truly groundbreaking. Next week they’ll report that most startups in the great game of venture-funded aura farming also don’t generate revenue, but at least those webshits get free LaCroix while failing. Hackernews shows up to miscorrect each other about market microseconds and immutable code, as if any of these abandoned smart-contract grave markers were ever more than automated token dispensers for early investors. The entire ecosystem is disguised unemployment for clankers and speculators, but sure, let’s pretend the scandal is that the ghost towns aren’t profitable.
The GPT-5 rollout has been a big mess
2025-08-11 | comments
OpenAI’s latest token shuffler face-plants onto the internet and everyone acts shocked that the world’s largest automated plagiarism engine can’t even roll out an update without detonating its own user base. The digital feudal lords blame a “broken autoswitcher,” which is Silicon Valley dialect for “we cut corners until the corners cut back.” Hackernews, ever eager to miscorrect each other, oscillates between pitying people who formed parasocial bonds with a clanker and insisting the company owes them emotional stability. Meanwhile the industry continues its favorite pastime: vibe coding trillion-parameter mood rings, then feigning confusion when users treat them like friends. The great game grinds on; the humanity gets smaller.
Crypto founder Do Kwon pleads guilty to US fraud charges
2025-08-12 | comments
Another crypto wunderkind discovers that the great game has consequences once the tokens stop levitating. Do Kwon, former “entrepreneur” (business model: Uber for setting billions on fire), finally pleads guilty, surprising exactly zero people except the Hackernews comment brigade still workshopping galaxy‑brain theories about whether he meant to commit fraud, as though intention matters when your entire career is automated financial arson. HNers dutifully turn sentencing speculation into a prediction market because of course they do; these people would turn their own funerals into Manifold bets. Meanwhile, the tech industry pretends this is an outlier instead of the inevitable result of giving webshits unlimited leverage, infinite confidence, and a clanker-powered hype machine to launder their vibes into currency.
Sam Altman says in decade, college grads will be working jobs in space
2025-08-12 | comments
Altman tells the proles that their grim future as algorithmically‑managed precarity goblins is actually a stepping stone to a glamorous career “exploring the solar system.” Sam Altman, digital feudal lord of the automated‑plagiarism‑engine economy, has apparently decided that what Gen Z really needs — besides school-loan forgiveness, a functioning labor market, some viable option for healthcare that is more than hope and prayer, and maybe electricity bills they can pay without auctioning organs — is reassurance that in ten years they’ll be packing their bags for Mars, no doubt propelled by the flames of the VC money that his company is currently burning.
Hackernews, ever eager to miscorrect each other into oblivion, spends 300 comments debating whether the media is unfair to the poor misunderstood billionaire, as though Altman’s entire public persona isn’t built on tossing out techno‑rapture slogans while quietly turning the internet into a slurry of vibe‑coded clanker spit. A few HNers bravely insist the quote is “nothingburger,” which is technically accurate only in the sense that everything about the current AI gold rush is aggressively bun, no meat.
Meanwhile, the rest of the thread disintegrates into the usual ritual: some defend the man because they read half a Paul Graham essay in 2009, others point out that telling a generation drowning in housing debt and collapsing job markets to “just go to space” is peak tech‑bro derangement. Naturally, no one mentions that the only people getting off this planet are the ones who already bought their ticket.
What does Palantir actually do?
2025-08-13 | comments
Palantir, the world’s most self-serious spreadsheet consultancy (business model: Uber for government panic-buying), has once again convinced a reporter to pretend its glorified ETL scripts are wizardry stolen from Elrond’s attic. The employees apparently call themselves hobbits, which is adorable if you ignore that Tolkien’s hobbits weren’t known for metastasizing surveillance states. Hackernews spends the thread miscorrecting each other about whether Palantir sells data, insight, magic, or just extremely expensive slide decks, while quietly drooling over the revenue-per-employee numbers like good little aspiring digital feudal lords. In the end, everyone agrees Palantir is basically Excel with a Pentagon interface, except much worse and somehow always the winning bidder.
AI Slop and the Destruction of Knowledge
2025-08-14 | comments
Another week, another academic discovering that the internet’s new favorite automated‑plagiarism‑engines are shoveling undifferentiated semantic landfill into every crevice of human knowledge. The article pretends to be shocked that Elsevier (business model: “Uber for paywalls”) — has quietly let clankers ghostwrite half their citations. Hackernews responds with its usual cocktail of sci‑fi fanfic and extinction fantasies, because nothing soothes a webshit’s ego like imagining humanity wiped out by the same vibe‑coding slurry they spent the last year copy‑pasting into production. Everyone performs amazement that turning the entire corpus of human writing into SEO slurry might have downsides. The Great Filter isn’t AI; it’s us, willingly drowning in our own trash.