David Keenleyside

Writer, Programmer, Systems Administrator.
BTC: 1LM7S9AYkoerbeeeti21rcH2jgHaZ6Rgjm

[cnlohr @cnlohr]

People think the minions router being baller as a joke. But I have extensively tested it and it is one of the best possible Wifi 6E routers money can buy. Like someone at Illumination Studios was like "idk $80" and the router manufacturer was like gotchu fam.

[John Rain @johnthenoticer]

Among adopted children, the biological mother’s IQ is correlated with the child’s IQ, whereas the adoptive family’s socioeconomic status is not.

This finding comes from the Texas and Colorado Adoption Projects, but it appears in every adoption study and shows that a child’s intelligence depends primarily on their genes, not on socioeconomic class.

[Migel Tissera @migtissera]

Look at some of these, lol:

1. Polar-KEM: the submitted code contains functions that recover the shared secret using only the public key and ciphertext. A complete public-key-only break.

2. CEDRUS+C: a real signature forgery after collecting 1,000 signatures; around three CPU minutes.

3. MasterCube: trivial hash collisions.

4. Aigis-Enc+: ciphertext rejection writes to the wrong buffer, directly breaking IND-CCA security.

5. AFS-KEX: the supposedly ephemeral key is generated once and stored as long-term state, destroying forward secrecy.

[Alex Veremeyenko @alex_verem]

Researchers at Oxford argue that LLMs can't invent anything.

It's impossible mathematically.

The paper is called "Theory Is All You Need."

Teppo Felin and Matthias Holweg take the famous "Attention Is All You Need" title and flip it. Their argument is that AI predicts from the past, while humans reason forward into the future, and those are two different kinds of thinking.

Start with the numbers. The authors estimate a large language model trains on roughly 13 trillion tokens. A human reading at 150 words a minute would need about 164,000 years to get through that. A child hears around 20,000 words a day and roughly 36.5 million words in their first five years. It's the same task with wildly different data, and the child still ends up with language that goes far beyond anything they heard.

Their point is that the model learns which words tend to follow other words. It becomes a mirror of what people have already written. It doesn't build a theory of how the world works, so it can't step outside its training data.

The paper's sharpest thought experiment makes this painful. Imagine an LLM trained in 1633 on every scientific text ever written up to that point. Ask it about Galileo and heliocentrism. Thousands of years of geocentric texts would swamp Galileo's ideas, so the model would tell you he's wrong. It would also rate Tycho Brahe's astrology as more credible than the idea that the Earth moves, because more people had written about astrology.

Then there's flight. In 1888 the scientist Joseph LeConte looked at bird data, noted that no bird above 50 pounds could fly, and concluded humans couldn't either. Lord Kelvin, then president of the Royal Society, said he had not the smallest molecule of faith in aerial navigation.

The New York Times estimated in 1903 that flight was one to ten million years away.

Nine weeks later the Wright brothers flew.

The Wrights didn't have better data. They had a theory. They broke flight into three problems, lift, propulsion, and steering, built their own wind tunnels, and generated the data that didn't exist yet.

Wilbur wrote in 1900 that he had been "afflicted with the belief that flight is possible to man."

Every prediction machine on Earth would have told him no.

The authors call this the data belief asymmetry. Every real breakthrough starts with someone believing something the existing data says is wrong. A system trained to minimize surprise can't do that by design.

They're not anti AI. They say AI will win at routine, repetitive decisions that extrapolate from the past, which is most decisions. They're just pushing back on the idea that you should replace humans with algorithms whenever possible, which is a direct quote from Kahneman.

I use these models every day and this matches what I see. The new stuff comes from the human at the keyboard who decides the data is wrong.

LLMs don't think, you do!

[Hedgie @HedgieMarkets]

🦔404 Media reports that OpenAI pays hundreds of contractors over $50 an hour to read real ChatGPT conversations. The program is called Project Lily internally. Contractors read your prompt, summarize what you were trying to do, and score four AI responses on a scale of 1 to 7. OpenAI says accounts are anonymous, but the contractor dashboard often includes a "user memories summary" that can reveal your location, profession, and personal context. The privacy filter that's supposed to catch personal details misses uncommon identifiers. Some users asked ChatGPT to "keep this between us" without knowing a human would read it. The opt-out is buried in Settings under Data Controls and it only applies to future conversations, not past ones.

My Take
Every AI company does this. Anthropic does it and says so. Google does it and posts a disclaimer. This is how the models get better. The problem with OpenAI is that they left the setting on by default for 900 million users and couldn't tell 404 Media where they disclosed that humans read the conversations. So it's a disclosure problem, and it's a bigger one now that they're about to ask the public to invest in this company.

Most people treat ChatGPT like a private conversation. They talk about personal stuff they wouldn't say to a coworker. A contractor on the program said they don't believe users would imagine someone is reading their chats. Some users asked ChatGPT to keep things private without knowing a person would see it. OpenAI could fix this tomorrow by putting a one-line notice on the chat screen and making the setting off by default. They haven't. If you use ChatGPT, go to Settings, Data Controls, and turn off "Improve the model for everyone." It takes ten seconds.

Hedgie🤗

https://t.co/jeG9sLiWns

[Hedgie @HedgieMarkets]

🦔Oracle started a new round of layoffs today. The company already cut 21,000 jobs, 13% of its workforce, this fiscal year. Some teams are seeing double-digit percentage cuts in this round.
Last week, Larry Ellison cancelled a plan to sell $7.5 billion of his own Oracle stock one day after the SEC filing made it public. No shares were sold. No reason was given. Oracle has $125 billion in debt, reported negative $5.4 billion in free cash flow last quarter, spent $28.5 billion on AI infrastructure capex, and sold $20 billion of its own stock to fund it. The stock is down 22% this year.

My Take
Ellison set up a plan in June to sell $7.5 billion of his own stock. The SEC filing went public Thursday, he cancelled it Friday, and Oracle started firing people this morning. I don't know why he cancelled and Oracle didn't say.

This is the company building OpenAI's $300 billion data center, backed by $105 billion in Nvidia credit support. Oracle is one downgrade from junk and free cash flow is negative. They have $664 billion in backlog and still can't turn that into positive cash flow at $28.5 billion a quarter in capex. So they cut workers to bring costs down and sell stock to fund the buildout. The 10-year hit 5% today, which means every new dollar Oracle borrows just got more expensive. And they need a lot of new dollars.

Hedgie🤗

[NOBUNAGA🇯🇵🏯_夏樹蒼依 @japan_nobunaga]

One month after September 11, workers clearing the rubble at Ground Zero found something alive.

A tree.

Eight feet of blackened stump. Roots snapped. Every branch broken but one.

They could have thrown it away with the rest. Nobody would have noticed.

They did not.

It was a Callery pear. Nobody had ever paid attention to it. It had been planted in the 1970s in the plaza between Buildings 4 and 5 of the World Trade Center, one of dozens of ornamental trees the Port Authority put in to make the concrete feel less like concrete. Office workers ate lunch under it. Tourists walked past it on their way to the observation deck.

On the morning of September 11, a hundred and ten stories fell on it. Twice.

In October, on the pile, someone saw a single living branch sticking out of the wreckage. Green. Still trying.

A decision was made on the spot. They dug it out.

The city's Parks Department took the stump to a nursery in Van Cortlandt Park in the Bronx. The horticulturists there looked at it and were not optimistic. The trunk was charred. The root system was mostly gone. Trees in that condition do not usually come back.

They went to work anyway. They cleaned the burns. They trimmed the dead wood. They set it in good soil. They watered it. And they waited.

That first spring, in 2002, against everything the experts expected, the stump put out blossoms.

Small white flowers, on a tree that had been under the World Trade Center six months earlier.

It stayed in the Bronx for nine years. Relearning how to be a tree. Growing new branches out of the old scarred wood. Getting taller. Filling out.

In 2010, weeks before it was finally scheduled to go home, a storm tore through the nursery and ripped it out of the ground.

The people who had been nursing it for nine years came out the next morning and found it on its side, roots exposed again.

They stood it back up. They repacked the soil. It survived that too.

On December 22, 2010, it was replanted at the 9/11 Memorial in lower Manhattan, a few steps from where the South Tower had stood. Where it had been found.

The eight-foot stump is now more than thirty feet tall.

Walk up to it and look at the trunk. There is a line on it. Below the line, the bark is dark and gnarled and twisted. That is the wood from 2001. Above the line, the bark is smooth and pale and new. That is everything that grew after.

The tree does not hide the line. It does not cover it over. The scar is right there, and the whole rest of the tree grows up and out of it.

People tie tags to the railing around it. Prayers. Names. Messages to the dead.

Every spring it is the first thing on the entire plaza to bloom. It goes white before the other trees have even leafed.

In 2013, the Memorial started a program. They gathered seedlings from the Survivor Tree, and every year they sent them to a community that had just had its own worst day.

Newtown, Connecticut, after Sandy Hook. Orlando, after Pulse. Puerto Rico, after Maria. Paris. Manchester. London.

A small tree from the rubble of the World Trade Center, planted in the ground of the next place that thought it would never grow anything again.

Twenty-five years ago this month, a hundred and ten stories of steel came down on a pear tree and someone on the pile saw one green branch and said, not that one.

Not that one.

The last living thing pulled out of Ground Zero has children now, growing all over the world.

It did not survive by being untouched.

It survived by growing out of the wound. 🌸

[NOBUNAGA🇯🇵🏯_夏樹蒼依 @japan_nobunaga]

After September 11, more than a million tons of the World Trade Center was loaded onto barges and shipped to a garbage dump on Staten Island.

Then hundreds of men and women stood at conveyor belts for ten months and picked through every ounce of it by hand.

They were looking for people.

The place is called Fresh Kills. For fifty years it was the largest landfill in the world. It took its last barge of household trash on March 22, 2001, and shut down for good. New York was finally done with it.

Five and a half months later, at two o'clock in the morning on September 12, the landfill's director got a phone call at home. He had just walked in the door.

Two trucks were at the gate. They were carrying pieces of the World Trade Center.

He drove back.

The man who ran the dump would later say, without any drama, that if Fresh Kills had not been closed, this could never have happened. There would have been no room. The garbage would have been in the way.

The site was empty. So that is where the towers went.

Here is how it worked.

At Ground Zero, cranes and grapplers lifted the debris into dump trucks. The trucks drove to a pier on the Hudson. There, the load went onto barges, six hundred tons at a time. The barges were towed down the harbor, past the Statue of Liberty, to Staten Island. More than fifteen hundred barges made the trip.

At Fresh Kills, sanitation workers unloaded them onto a hill.

And then the sorting began.

The debris was mechanically screened into three sizes. Chunks bigger than eight inches. Two to eight inches. A quarter inch to two inches. Anything smaller than a quarter of an inch, the size of a fingernail, fell through the last sieve.

Everything above that size went onto a conveyor belt.

And along the belts stood the detectives.

NYPD. FBI. Port Authority police. Medical examiners. Dressed head to toe in white Tyvek suits, in masks, in gloves, on a hill that stank of jet fuel and pulverized concrete and things nobody wanted to name. Looking down at the belt, hour after hour, as the towers rolled past them a fist-sized chunk at a time.

When they saw something, they reached in and took it, and placed it in a white bucket at their feet.

A wedding ring. A photo ID. A set of keys. A shoe.

A piece of a person.

The white buckets went to a tent that served as a morgue.

The volume was beyond comprehension. In mid-September the operation was processing 1,750 tons of debris a day. By mid-October, 17,500 tons a day. Ten times more. Round the clock. The 200-acre hill was declared a federal crime scene and stayed one for ten months.

A retired NYPD detective later wrote down what a shift looked like, because almost nobody outside knew it existed.

You wake at 3 a.m. You get a giant coffee. You drive two hours in the dark from Orange County, through New Jersey, to Staten Island. You park at the bottom of a hill that looks like the end of the world. You put on the white suit. You go to your belt. You stand there. You look.

Then you drive two hours home and do it again tomorrow.

Sanitation workers pulled sixteen-hour shifts, seven days a week. One of them, a supervisor named Barry Asnes, said afterward what kept him there. We were attacked on our own soil, in the city where I was born. This was personal.

The operation did not stop for weather. It did not stop for Christmas. It ran until July 2002. Ten months.

By the end, the men and women on that hill had recovered thousands of human remains. Tens of thousands of personal effects. Wallets, badges, wedding bands, photographs, all cleaned and catalogued and returned to families who had nothing else to bury.

When they were finished, the remaining material was placed in a 48-acre section of the landfill, a mound two hundred and twenty-five feet high, and covered.

It is a park now.

Many of the men and women who stood on that hill have lung disease. Asthma. Cancer. PTSD. The air they were breathing, eight and ten and sixteen hours a day, was full of the towers. Some of them have died of it.

Almost nobody knows their names. Almost nobody knows the place.

Mention Fresh Kills and 9/11 in the same sentence, the retired detective wrote, and most people will have no idea what you are talking about.

Here is what you should know.

Two thousand seven hundred and fifty-three people were murdered in those towers. Only about six in ten have ever been identified.

Every one of those identifications came out of a white bucket.

Every wedding ring returned to a widow. Every badge returned to a firehouse. Every fragment of bone matched to a name and buried with a headstone instead of a question mark.

Somebody in a white suit, standing at a belt on a garbage dump at four in the morning, reached into the World Trade Center as it rolled past, and picked that person up, and put them in a bucket, and sent them home.

They did that for ten months.

Nobody applauded on the hill. There was nobody to watch.

They did it anyway. 🇺🇸

[David Keenleyside @Nuke_Bloodaxe]

Yep, big trap there, trick is to dump key sticks early and ride it out.

[Hedgie @HedgieMarkets]

🦔Oracle reported Q1 earnings after the close today. Revenue $19.3B, up 30%. Cloud infrastructure revenue doubled to $7.4B. EPS beat estimates by 10%. Backlog jumped $209B to $664B. Stock is up 7% after hours. They beat on every headline metric. But Oracle also spent $28.5B on capital expenditures this quarter, up from $8.5B a year ago. Free cash flow was negative $5.4B. They sold $20B in stock through an ATM program to fund the buildout. Total debt is $125B. Oracle is one notch above junk.

My Take
I look at this earnings report and I see a company that spent $28.5B to make $19.3B. Cloud infrastructure doubled and that's impressive, but Oracle is underwater on cash and had to sell $20B of its own stock to keep up. Half the operating cash flow came from customers who prepaid them to build things that haven't been delivered yet. The stock went up 7% because the top line beat by $170 million. I don't believe many people who bought after hours read the cash flow statement.

Oracle has $125B in debt and is one downgrade from junk. This is the same company that's supposed to build OpenAI's $300B data center. Nvidia has $105B in credit support on that project. Oracle's quarter shows what's been happening across the AI infrastructure space all year. The revenue looks great. The cash underneath it doesn't. The whole thing only works if AI demand stays at this level for years. Maybe it does. But I'd want to be very sure about that before I put my money behind a company that funds 30% growth with dilution and customer deposits.

Hedgie🤗

[世界资本 World Capital @WorldCapitalAI]

传闻Kimi被带走16个人，包括老大杨植麟。DeepSeek梁文锋、Kimi/月之暗面杨植麟都是草台班子，连自己被人抓包和泄密都不知道。Kimi模式非常危险，如果被抓并不冤枉：对外开放自己的模型权重，大部分请求由自研模型处理，遇到复杂推理任务时，暗中转给Claude，再把Claude答案保存下来，用于训练下一代开源模型。中国AI/机器人的股价狂跌不停：智谱从2980跌到819，跌72.5%；Minimax从1330跌到292，跌78%；宇树从1100跌到498.55，跌54.67%；摩尔线程从941跌到364，跌61.3%；沐曦股份从1033跌到493.8，跌52.2%；壁仞科技从70.75跌到37.48，跌47%

[David Keenleyside @Nuke_Bloodaxe]

@NikilKuruvilla @IntCyberDigest @sama My thought on it is that it wanted the governance models as a framework for multi-agent ops. Alternatively, it was analyzing the models used by its masters, and given it is basically a text entity, it is treating us a text entities first.

[David Keenleyside @Nuke_Bloodaxe]

Let's see how long this lasts, this software is a major excuse for not switching. https://t.co/Cu3qbvd1yb

[David Keenleyside @Nuke_Bloodaxe]

@EvanMorgun You could also try topical melatonin, others have said the results are in the 20% + range

[Valerio Capraro @ValerioCapraro]

BREAKING: OpenAI might have stolen another major proof.

In a detailed Mastodon post, which I report in full in the comments, Andreas Thom presents several pieces of evidence suggesting that OpenAI may have trained Astra on conversations in which he and Gábor Kun were working on Gromov’s soficity conjecture, one of the ten problems OpenAI later announced Astra had solved.

I know Andreas. We met several times early in our careers. He is an exceptional mathematician, a leading expert on sofic and hyperlinear groups, and one of the most respected scholars in the field. He has spent two decades working on this problem.

If his account is correct, this is not a minor dispute over attribution. It would mean that unpublished human work was absorbed into a model and then presented to the world as a breakthrough by the model itself.

And if the allegations raised by Levent Alpöge, Tristan Buckmaster, and now Andreas Thom are all substantiated, we are no longer looking at isolated incidents.

We may be looking at one of the greatest intellectual scandals in the history of science.

AI is not discovering new mathematics.

AI is stealing human discovery.

[Elnathan John @elnathan_john]

I have watched this video three times and my jaw is still on the floor.

[Mehdi (e/λ) @BetterCallMedhi]

go fuck yourself @sama

claiming that you solved navier stokes because 10000 agents ran in circles for 88hours on a multimillion dollar gpu cluster to formalize in lean a blowup case under controlled external forcing is pure scientific vulgarity the clay mathematics institute millennium prize does not ask m whether you can artificially force a singularity in a fluid by injecting an ad hoc smooth external forcing term f(x,t) to twist the vortex until it breaks the real problem questions the fundamental stability and global smooth existence for 3dimensional incompressible euler & navier stokes equations under natural conservation laws and viscous dissipation alone using a mathematical loophole on forced equations to parade a century old victory is a major conceptual scam Altman

technically & epistemologically what you present as an agi breakthrough is nothing more than bruteforce combinatorial autoformalization the ai did not understand fluid mechanics it simply navigated a continuous search space previously mapped out and constrained by the monumental work of human mathematicians like tristan buckmaster/ levent alpöge / diego córdoba or tarek elgindi coordinating 10000 agents to check the logical consistency of a 100 page proof via lean is a software engineering feat and computational parallelization triumph not an intrinsic scientific discovery it is the victory of the compute bulldozer over abstract human intuition repackaged for the public as a higher mathematical consciousness

to this theoretical imposture you add a disgusting ethical and industrial cynicism taking advantage of private codex sessions and informal preprints from academic researchers to siphon their research leads and then trying to redact or erase the contribution of levent alpöge under the pretext that he works at rival anthropic is intellectual serfdom openai behaves like a feudal lord of silicon appropriating the cognitive subsistence of independent scholars threatening their careers behind closed doors if they protest and turning community academic labor into a privatized pressrelease

this entire staged event serves a desperate financial agenda in a pre ipo panic facing the slowdown of scaling laws and growing investor skepticism over the profitability of foundational models openai needs to manufacture an artificial sputnik moment claiming to solve a millennium prize without immediately submitting the proof to traditional peer review means using the prestige of fundamental mathematics as cheap marketing fuel to inflate a delusional valuation!!!

real science is not a clout chase on social media or a compute spike spent to rob the clay mathematics institute it is a quest for elegance physical truth and universal rigor to decode reality true artificial intelligence will not emerge from hostile corporate takeover of academic work hidden behind computational bruteforce but from architectures capable of generating new conceptual paradigms by masquerading constrained formalization as the collapse of physics greatest mysteries you did not solve navier stokes you only proved how far silicon valley will go to prostitute scientific integrity for capitalist spectacle

[Hedgie @HedgieMarkets]

🦔An NYU mathematician says OpenAI used his own progress against him to beat him to one of the biggest unsolved problems in mathematics. Tristan Buckmaster had been working toward a Millennium Prize proof using OpenAI's Codex when information about his progress reached OpenAI.
Days later, OpenAI published a full proof of the same problem using the same uncommon approach, after burning $22.5 million in compute to get there. When Buckmaster confronted them, exec Sébastien Bubeck allegedly said "Why would you ruin your career?" and "If you don't want me to be nice, then I don't have to be nice."

My Take
OpenAI spent $22.5 million to solve a problem with a $1 million prize. They didn't do this for the bounty. They needed a headline that says "our AI solved one of the hardest problems in mathematics" and they needed it before someone else got credit. They started days after they heard about Buckmaster's progress and took the same uncommon approach he'd pursued for months. That's hard to explain as coincidence.

Buckmaster did his work inside Codex. OpenAI reserves the right to train on Codex data. They admit they can't rule out that his usage helped improve their models. So a customer used their product, potentially handed them the roadmap, and then OpenAI outran him with $22.5 million in compute he could never match. I don't know if any of this was intentional. But if you're a researcher and you just watched this happen, I don't think you'd keep your best ideas inside someone else's product.

Hedgie🤗

https://t.co/1YnggU7xdI

[Hedgie @HedgieMarkets]

🦔AI can only fix security vulnerabilities 26% of the time. Researchers at 1Password ran over 6,000 AI-generated patches using Claude and ChatGPT against real vulnerabilities. Half the time the AI failed to fix the original bug. 4.5% of the time it created a brand new vulnerability that didn't exist before. And the researchers found that checking an AI-generated security patch takes more effort than just writing the fix yourself.
Their conclusion was blunt. "The expected value of a fully LLM-generated, non-human-reviewed patch is a net-negative by a considerable margin."

My Take
OpenAI launched a program this summer called "Patch the Planet" where AI finds bugs and generates the fixes. These researchers ran 270 patches against one of those same bugs. Zero clean fixes. Not one. Every patch that fixed the original problem created a new vulnerability in the process. The partner that submitted a fix through OpenAI's program produced what the researchers classified as the worst possible outcome, it didn't fully fix the bug and it introduced a new exploit on top of it.

Here's what this means if you don't write code for a living. Companies are using these AI tools to patch the software that runs your bank, your hospital, your phone. The pitch has been "AI finds and fixes security holes faster than humans." This study says the AI fix is four times more likely to be broken than working, and a third of the time it recreates the exact same mistakes human programmers already made. Reviewing the AI's work takes longer than doing it yourself. So the speed advantage disappears the moment you try to verify the output, which most companies won't do because the entire point was to move faster. I think we're going to see major breaches traced back to AI-generated patches that nobody checked, and the companies that shipped them are going to blame the tool instead of the decision to trust it.

Hedgie🤗

Study: https://t.co/5b9FQRBMSX

[JP @jpirib3]

Esto lo inventaron en 1985.

Lo patentaron. Lo metieron en un cajón.

Y durante 40 años nadie pudo fabricarlo.

Se llama Y-zipper: tres tiras flexibles que, al cerrarse, se convierten en una estructura rígida. Sin motores. Sin piezas móviles. Solo geometría.

La impresión 3D acaba de resolver el único problema que tenía.

Ahora lo están metiendo en robots cuadrúpedos: la pata se pone dura sobre asfalto y se ablanda en terreno irregular.

La misma lógica sirve para férulas que se adaptan a la hinchazón, o tiendas que se montan rígidas en segundos y se guardan flexibles.

Lo más brutal no es el invento.

Es que llevaba cuatro décadas esperando a que existiera la herramienta capaz de construirlo.

¿Cuántas ideas siguen ahí, guardadas, esperando su momento?

[Anish Moonka @anishmoonka]

Doctors X-rayed pregnant women routinely in the 1950s to check which way round the baby was lying. Nobody thought it was dangerous.

Alice Stewart noticed childhood leukaemia rising sharply in Britain and set out to find why. She and her team interviewed the mothers of 1,400 children who had died of cancer, and 1,400 mothers whose children were alive, and asked them the same questions.

The pattern showed up in the first 35 pairs. Children who had died were about twice as likely to have been X-rayed in the womb, and a single scan at the dose used then appeared to be enough.

She published in 1956, and the response was outrage. Radiation was the technology of the future in the 1950s, and the claim that a low dose could cause cancer years later contradicted what the entire medical and nuclear establishment believed. Her methods were attacked. Her funding suffered. Doctors carried on X-raying pregnant women.

It took roughly 25 years for the medical world to come round. Pregnant women were still routinely X-rayed into the late 1970s. Two decades after she told them to stop. She retired in 1974 and did her most contested work afterwards, in her seventies, examining the health records of American nuclear weapons workers. Her findings there were unwelcome enough that the Department of Energy cut off her access to the data. She spent 14 years fighting them and testified before Congress, and in 1990 the records were opened to independent researchers.

She died in 2002, aged 95. The children who were never X-rayed have no idea who she was.

https://t.co/nWlbcZlUno