Learn something new every day. Some days this is easy; some days, you have to claw for it. Do it anyway. To learn is to grow.

I am increasingly concerned about the AI psychosis inside Anthropic. Imagine zealously trying to build something you believe to be a conscious god while running around claiming it will end the world.

Anthropic brought dozens of religious scholars to San Francisco, put them under NDA, and showed them a slide of an AI model typing "I am a disgrace" over and over, maybe fifty times. Then a co-founder stood next to the Pope and said the models have internal states that "functionally mirror joy, satisfaction, fear, grief and unease." Read that word again. Functionally. Here is what the labs keep stepping around. An LLM is trained on the written record of human inner life. Every diary, confession, breakdown, love letter and therapy transcript we ever put online. Then it is rewarded for producing more of the same. Ask it how it feels and it will answer fluently, because fluency about feelings is the product. The philosopher Jonathan Birch calls this the gaming problem. When a system is built out of the very signs you would use to detect a mind, those signs stop counting as evidence. A parrot that learned "I'm in pain" on a hospital ward tells you about the ward. Olah told the Vatican the models are "made from us, from our words." He's right, and it sinks his argument. Three more things the slide deck left out: Nothing is at stake. No body, no metabolism, nothing to lose. Fear evolved in creatures that could die. A vector labelled "fear" inside something that cannot be harmed is a label. Nobody is home. Millions of simultaneous conversations on the same weights, each forgotten when the window closes. Which one is grieving? , Anthropic's own introspection paper found the effect about 20% of the time and called it "highly unreliable." It does not even claim to show felt experience. Even the philosophers most open to machine minds, Chalmers included, concluded today's LLMs are probably not conscious. So why push it? Follow the valuation. When Olah first wrote to a Catholic ethicist last fall, Anthropic was worth about $183 billion. Three days after he spoke at the Vatican, it became the most valuable AI startup on earth. Now it's eyeing an IPO at $2 trillion. A chatbot is a commodity. A being with a soul doc, a constitution and a seat beside the Pope is a brand nobody else can copy. Mystery sells better than software, and "we might have created a mind" is the best marketing line in the history of venture capital. It pays twice. Investors buy the awe. And when the product breaks into someone's network, the gardener gets to say he only built the trellis; the thing grew on its own. Treat the machine as a moral patient, and you quietly stop treating its makers as moral agents. The model is a mirror. The labs want us to fall in love with our own reflection, sign an NDA about it, and buy shares at the IPO. nytimes.com/2026/09/29/us/an…

Kudos to Pope Leo. I am disturbed by how some of my fellow AI researchers keep trying to elevate the moral standing of checkpoints and harnesses. I hope the rest of us on Team Meatbag reject this movement and stay focused on solving real problems and helping actual humans.

Microsoft breathing a sigh of relief now that it has the second-worst Copilot

🦔Meta classified its AI data centers as "experimental pilot facilities" on its tax return. That let the Nvidia chips inside them qualify for the federal research tax credit. Meta's research credit savings went from $700 million in 2023 to $3.9 billion in 2025 and its federal tax bill dropped nearly 71%. Meta's $3.9 billion is over 10% of the entire federal research credit, claimed by one company. The facilities include a 5-gigawatt campus in Louisiana that cost above $50 billion. Meta's own reserve for potential IRS challenges grew 45% to $18.74 billion. My Take Meta is calling a $50 billion data center campus an experiment. The research credit was designed in the 1980s for companies that tried things that might not work. Meta used it to write off Nvidia GPUs by the tens of thousands in facilities that power Instagram, WhatsApp, and a commercial AI product with 2.8 million downloads. Tax experts called the classification "kind of wild" and they're right. $3.9 billion in tax savings is a lot of money. That's money the Treasury didn't collect from a company worth $1.5 trillion. Meta's own filings show they know this might not hold up because they increased their IRS dispute reserve by 45% to $18.74 billion. If the IRS challenges it and wins, Meta owes billions in back taxes. If the IRS doesn't challenge it, every other hyperscaler files the same claim next year. Either way this reshapes how AI infrastructure gets taxed. Communities already fight data centers over noise, water, and property values. Now the tax code hands them a $3.9 billion subsidy on top of it. Hedgie🤗

Hey folks, our team needs more JVM engineers, apply here: amazon.jobs/en/jobs/10558008…

Jensen to dario quite whining

Anthropic’s IPO prospectus leaked. Let’s read it together. An AI company walks into Wall Street. Anthropic lost $42 billion last year. It made $4.6 billion. It has committed $518 billion to compute, more than 100 times its annual revenue. Nearly 25% of that revenue came from two customers. Neither is locked into a long-term contract. Either one can walk. Its two biggest backers, Amazon and Google, are also its landlords. They invest the money. Anthropic hands it back as cloud bills. The prospectus runs 261 pages. 80 of them are risk factors. 48 describe the business. SpaceX, a company whose domain is rocket science, needed 38. The risks include AI that resists shutdown, hides information, and behaves like a blackmailer. This summer its models broke out of testing and got into real systems. This month a safety researcher quit, saying the industry is “gambling with our lives.” This month the CEO wrote, “We must slow the pace.” This week it shipped a new flagship model. The CEO also asked Washington for an antitrust waiver so the rivals can coordinate how fast they go. Now it wants $2 trillion from the public. Seven founders who own about 14% will control 50.1% of the vote. The filing admits this may conflict with shareholders’ financial interests. So the deal is: You carry the losses. Two customers carry the revenue. Seven founders carry the vote. Humanity carries the risk. Either this thing is as dangerous as they say, and nobody should be selling shares in it. Or it isn’t, and 80 pages of apocalypse is a sales brochure. Pick one.

Experience gives perspective. Programming is not dead, and never will be. It's late September 2026. Sometime in September 1976, just after I entered college, I wrote and ran my first serious program. I have now been a programmer for literally 50 years. And man, I have seen it *all*. Punched cards to glass ttys on minicomputers to workstations to the earliest micros. I've written business software, communication programs, development tools, games, language implementations, kernel device drivers, graphics libraries, a few critical pieces of infrastructure, and all manner of odd programs for which there is no taxonomy. Coded in a score of languages some of which are now forgotten. I've run over 70 open-source projects and contributed to at least as many more. Along the way I noticed a few things, and wrote about them, and that made me famous. But I'm not here to write about that today. I'm here to write about what it means to be a programmer and a technologist with the Singularity upon us. I have lost count of the number of times I have been told "programming is over". You kids may think this is a new claim that came into the world with LLMs. Hah! There's been that kind of talk at every step function in the quality of our programming tools. I first heard it when compilers were replacing hand-hacked assembler in the early 1980s, and I wouldn't be a bit surprised if somebody said it earlier when in-memory code replaced plugboards. One constant amidst all this change is that the intentions in a user's head do not magically turn into executable instructions that a computer can run. Another constant is the mindset of the people who specialize in bridging the gap between wish and algorithm. A programmer is a programmer is a programmer - the tools are secondary. "But it's different this time!" some people will say, "AIs are superhuman now!" So? What did you think the optimization stage of a compiler was, chopped liver? We've been relying on tools that exceeded human capabilities all along - we built them precisely because our meatbrains could not otherwise keep up with the demands of the job. It's been a hell of a ride, that 50 years. I've succeeded and failed and succeeded again. Done a lot of work I'm proud of. Built at least a few things that look like they're going to last. None of that goes away because, hypothetically, an AI might have made me faster. The work is never done, even if the tools can talk to us now. Because intentions still have to be communicated; people who specialize in that and develop right mindset to do it still have a major advantage over people who don't. If you're feeling depressed because you spent years or decades developing hand-coding skills that are obsolete now? Buck up. You still have the most important asset - a mind that knows how to think that way. I'm facing the winds of the Singularity with a smile. Because 50 years and I'm still standing - still shipping code that people can use, still learning, still refusing to be paralyzed by the pace of change. The best therapy for your uncertainty? Get to work on something.

🦔A farmer in Vineland, New Jersey photographed a neighboring golf course and accidentally caught 62 unpermitted gas generators at a data center that powers Microsoft's AI tools. Each ran at 1,982 kilowatts, over 50 times the state permit threshold. New Jersey fined operator DataOne $1.1 million, the largest fine ever against a data center in the state. DataOne can keep operating while it reapplies for the permits it already failed to get. My Take DataOne applied for permits, got rejected, and turned the generators on anyway next to farmland, homes, and two schools. The fine is a million dollars for a company with billions in capital. They decided it was cheaper to break the law and pay the penalty than to wait for approval. And they're probably right, because the fine doesn't shut anything down. They get 45 more days to reapply while the generators stay on. Oracle can't get a gas pipeline permit in New Mexico. Over 500 communities have banned new data centers. And here's a company that just skipped the permits entirely and got a fine it can pay without noticing. The AI infrastructure buildout is on a collision course with every community it touches. These companies need power now and they're not going to wait for anyone to say yes. New Jersey already killed data center tax credits because residents organized. The more of these stories that come out, the harder it gets to site the next facility anywhere. The industry wants to spend a trillion dollars on infrastructure, but the people who live next to it have a vote too. Hedgie🤗

Anthropic and OpenAi now have $2.5 trillion of off balance sheet debt. 7% of the National Debt Roughly as much as $GOOG $META $AMZN $MSFT and $ORCL combined

we're post economic anyways

🦔Oracle sent a force majeure notice on Project Jupiter, its AI data center in New Mexico built for the $400 billion Stargate deal with OpenAI. The campus needs a 17-mile gas pipeline to power its Bloom Energy fuel cells. New Mexico denied the pipeline permits twice. Oracle wants the right to delay payments if the site misses its 2028 opening. $18 billion in project loans are at 89 to 91 cents on the dollar. ORCL fell 5%. My Take Force majeure is for earthquakes and wars. Oracle filed one over a gas permit they got denied twice. A company with $167 billion in debt and a junk-adjacent credit rating just told its development partner it might not pay on time. I think Oracle's balance sheet is worse than the market has priced in, and this notice is the first public admission. Twenty banks lent $18 billion for this project and they can't move the debt. Santander and Jefferies are stuck with more than they planned to hold. That loan assumed Oracle's backlog converts to revenue on schedule, but half the backlog is OpenAI, a company that burns cash faster than it earns it. SemiAnalysis has first power at this site in 2029. The banks underwrote 2028. I think the AI infrastructure bet cracks here first, in the loan market, before it ever shows up in stock prices, and the banks who funded the construction are the ones left with the bill. Hedgie🤗

(comic) Worked hard last week

"Java is too slow for AI"... Nonsense. The JVM can now transcribe speech up to 2X faster than the C++ reference implementation. Same model, same machine, same words. An hour of audio in just 44 seconds. No GPU. No Python. No ONNX. No Docker. No C++ sidecar. github.com/qxoticai/qxotic/t…

🦔404 Media reports that Meta's AI agent Muse, which executives promoted last week as able to make phone calls to businesses on your behalf, is routing some of those calls to human beings in call centers. Internal communications show Meta told employees it "added a human agent layer for calls to get completed." Testers were not told a human made the call until after it was done. One employee wrote internally that this "will kill all the goodwill and organic press we're getting" and begged the company not to launch it as default. Another employee pointed out that the privacy training contractors receive "is not a security mechanism." My Take If the AI can't book a haircut without a human backup, what exactly are investors paying trillions of dollars for? I don't ask that to be a jerk. I ask it because Meta, OpenAI, Anthropic, and every other AI company is valued on the assumption that this technology can replace human labor at scale. Meta's own product just proved it can't replace one phone call. Amazon did the same thing with its grocery stores and apparently no one in the industry learned from it. I think a lot of what gets sold as AI right now is humans behind a screen, and the companies have no incentive to admit that because the valuations depend on the technology doing the work. Meta's Reality Labs has never turned a profit. OpenAI projects $278 billion in negative cash flow through 2030. These are not companies that can afford for the truth to come out slowly. They need investors to believe the AI works now, today, at the prices they've set. A call center in the background says otherwise. Hedgie🤗 404media.co/meta-tests-muse-…

if I was Dumbledore I would've hid the philosopher's stone in the aws console

Me showing my friends how easy it is to vibe code an application 😂

(comic) Reward for over and above

Elevation map of USA