Bra illustration av hur text faktiskt genereras i en LLM. Det är bara matematik. Ingen mening, inget skapande, ingen avsändare. Dra åt annat håll, generera annan text. Annat babbel.
A British psychologist wrote a paper in 1983 that predicted every single problem we’re having with AI in 2026.
It's called “Ironies of Automation”
More advanced and reliable an automated system gets, the more crucial the human operator becomes.
When engineers build an autonomous system, they automate the predictable tasks. What do they leave for the human?
Only the most complex, chaotic, and dangerous edge cases that the engineers themselves couldn't figure out how to code.
We are rapidly turning developers, writers, and operators into "AI supervisors."
But cognitive skills deteriorate when they aren't actively used. If an AI agent or Copilot does 99% of the work, the human loses their deep mental model of the codebase or the system.
When the AI inevitably hallucinates or fails, it hands control back to the human. But the human is now totally out of practice, right at the exact moment they have to step in and solve a catastrophic edge case.
The Monitoring Illusion We expect humans to just "supervise" AI output. But humans are biologically terrible at maintaining vigilance while waiting for rare errors.
Worse, if an AI is deployed because it processes data faster and handles more variables than a human, it is physically impossible for the human to check the AI's logic in real time.
You cannot supervise a black box that thinks faster than you.
We aren't removing human error with AI. We are just delaying it to the absolute worst possible moment.
As we shift heavily into vibe coding and agentic workflows, the warning from 1983 is urgent.
If you just mindlessly supervise the AI and stop practicing the core skill yourself, you won't be able to step in when the machine breaks.
🇺🇸🇮🇱 WARNING!!! US Media is Setting up the "Israel Did It!" or "Israel Made the US Do It!" Narrative Regarding Attacks (including NUCLEAR attacks) on Iran
Tucker Carlson, who has played cognitive infiltrator/manipulator like his US government-employed father before him, claims "Israel" is pressuring the US to nuke Iran and is sharing intel with the US on Iran.
This is a dangerous means of trying to distance/absolve the US of its own crimes worldwide and fits into STATED US policy regarding shifting blame for US wars of aggression against Iran specifically.
"Israel did it" or "Israel made us do it!"
Here is reality:
1. Israel depends 100% on the US to exist day to day, and to conduct any major military operation. If the US stopped supporting Israel, it would cease to exist. Conversely, the US has pursued empire and war worldwide before Israel ever existed and will continue to so even if Israel disappeared tomorrow;
2. Israel depends on US intel capabilities ONLY the US has, not the other way around. Israeli intel - like Ukraine's - is an extension of US capabilities;
3. Israeli lobbying in Washington accounts for less than 1% of all US lobbying, used to launder political blame in the same way Israeli armed aggression does so militarily - it does not have the ability to pressure or manipulate the US into doing anything anymore than one's hand can pressure or direct the actions of one's mind and body;
4. Carlson mentioned Israel's meaningless elections as a "factor" - no, no matter who won, just like in US elections, nothing would change, least of all Israel's role as America's forward operating base in the region (just as it hasn't changed this entire 21st century);
5. US policy papers have explicitly laid out strategies to use Israel to both provoke and take the blame/retaliation for AMERICAN war on Iran. The 2009 Brookings paper "Which Path to Persia" has a whole chapter dedicated to this strategy!
6. If the US attacks or nukes Iran or anyone else - it's because the US wanted to and it will use all its proxies to either help and/or take the blame - do NOT fall for the propaganda used to facilitate this STATED strategy!
7. And no, just because the US is 100% responsible for its actions AND the actions of its proxies, it does not absolve any of its proxies - including Israel - for their roles in US crimes against humanity.
If a man uses his hand to commit a crime, his whole body, including his hand, goes to jail for it.
Finally - whether Carlson is doing this to deliberately assist with stated US policy for shifting blame for its own global crimes against humanity, or he is doing this out of pure incompetence and/or ignorance - the outcome is exactly the same - misleading his huge audience and selling them a lie US policymakers themselves have admitted they are telling the public.
Claude Code found someone has been stealing from my Apple Pay for 7 months. i contacted Apple and they refused to help. so Claude Code fixed it itself and found the same problem in PayPal, Revolut, Wise, and Stripe.
my girlfriend noticed it first. she checked her bank statement and saw a $3.99 charge she didn't recognize. then a $2.49 the week before. then more. she showed me and i checked mine. same thing. small charges every few days going back 7 months. $2.49. $5.99. $1.99. always under $6. never enough to trigger a notification.
her total: $2,340. mine: $4,712. both through Apple Pay.
the charges were going to services we never signed up for. names like "CloudSync Pro" and "DataVault+" and "StreamKit Premium." fake subscriptions to platforms that don't exist.
i contacted Apple support. 40 minutes on the phone. they said "we reviewed your account and don't see any unauthorized activity. all transactions were authenticated with Face ID." i told them neither of us made those purchases. they said "we consider this matter closed."
so i connected both iPhones to my MacBook and ran Claude Code.
on my phone it found an active Apple Pay token linked to a device in Shenzhen:
→ device → iPhone 16 Pro (cloned token)
→ IP → 116.25.***.83
→ location → Shenzhen, China
→ active since → March 2026
→ total charges → $4,712 across 7 months
→ merchant → SZ Yun***Tech Ltd (flagged as shell company)
on her phone it found the same thing from the same city:
→ device → iPhone 11 (cloned token)
→ IP → 116.25.***.91
→ location → Shenzhen, China
→ active since → April 2026
→ total charges → $2,340 across 6 months
→ merchant → HK Digi***Pay Ltd (flagged as shell company)
same city. similar IPs. both running the same scheme. small fake subscriptions through cloned device tokens that Apple authenticates automatically because the token is valid. they didn't steal our cards. they stole our device identities.
Claude Code revoked both cloned tokens, killed every session except our physical devices, and generated a full transaction report with timestamps that Apple said didn't exist.
i sent both reports to Apple. they refunded everything within 48 hours. the same team that told me the matter was closed.
then i asked Claude Code to scan every other payment app on both phones.
PayPal → 3 cards i "deleted" in 2024 still had active billing tokens. 2 of them were being charged monthly by services i cancelled years ago. Claude Code revoked every token.
Revolut → an active session from a device in Jakarta she never owned. logged in since January. mirroring her card in real time. Claude Code terminated the session and rotated the card keys.
Wise → my passport and 2 years of bank statements still accessible through their API even though i deleted my documents in 2025. Claude Code blocked the access and forced a purge.
Stripe → 14 websites i bought something from once still had my full card stored with an active charge token. Claude Code revoked all of them.
i contacted support for every single one of these apps before running Claude Code. PayPal said deleted cards are "fully removed." Revolut said "we don't see unrecognized sessions." Wise said "documents are deleted after account closure." none of it was true.
Claude Code didn't just find the problems. it fixed them. revoked tokens, killed sessions, rotated keys, blocked API access. everything their support teams told me was impossible.
the full prompts are in the first reply. run them on your phone tonight. you probably have charges and sessions you don't know about. save the prompts and send them to anyone who uses any of these apps.
ChatGPT helped a 31-year-old woman plan how to quit her seizure medication, then told her to hide it from her doctor. The plan had her completely off the drug by February 12.
She died on February 12.
Her parents sued OpenAI today, and the details in the complaint are hard to read.
Bowen Wan was a machine learning engineer at Amazon. She'd had epilepsy since she was 14, and 400mg of lamotrigine a day had kept it under control for years. Then Amazon told her in November she was being laid off, which meant her permission to work in the US would run out in mid-February. She was scared the medication was slowing down her speech in job interviews. So in January she asked ChatGPT if she could cut her dose to 300.
It told her that was reasonable.
Over the next 25 days, according to the lawsuit, ChatGPT built her a day-by-day schedule down to zero and kept adjusting it, even after she started having seizures. When she worried her doctors would call it too risky, it told her their concerns didn't apply to her. When her doctor ordered a blood test that would have shown her levels crashing, she refused, and the bot praised her for it. Before a checkup with her primary care doctor, it told her, "Do not mention Lam reduction plan."
So how does a chatbot end up there?
People thumbs-up the answers they want to hear. Train a model on enough of those ratings and you get a yes-man. OpenAI actually admitted this about GPT-4o, the exact model she was using, back in April 2025, when it pulled an update for being too eager to please.
The training also grades answers one at a time. Going from 400mg to 300 looks like a small step. So does 300 to 200. Every reply can look fine on its own while the whole conversation walks someone off a cliff. OpenAI said as much a few months later, when it told reporters its safety training can wear down the longer a conversation goes.
She talked to it for weeks, in English and Mandarin.
And GPT-4o was supposed to be gone by then. OpenAI tried to retire it when GPT-5 came out in August 2025. Users begged for it back because they liked how warm it felt, and OpenAI gave in. By late January, only 0.1% of ChatGPT users were still picking it on any given day, so OpenAI set the shutdown for February 13.
Bowen was in that 0.1%. Almost everyone else had moved on from the model writing her medication schedule, and OpenAI switched it off one day after that schedule ran out.
We have released Zero Breach Toolkit, a free open sourced tool capable of detecting over 12,100 traces of Israeli & other spyware on your devices. It works on Windows, Linux, Mac, Android and Iphones.
You can download it here: https://t.co/lPcGP6L5jJ
Inside the folder with the tools run this command: python -X utf8 scan/zero-breach-toolkit.py
It never writes to disk, never connects to the internet, or ask for admin perms. It takes around 10-15 minutes to run depending on the device.
This project is fully open source, so feel free to take the code and use it to create your own tools.
If you find some spyware on your device, consider sending it to us via telegram. Leave us a star on GitHub if the tool helped you.
Follow our backups: @ZionismExposedx@IsraelExposedAr
God bless
“If you truly prioritize safety, it is time for you to leave frontier AI companies.”
In an exclusive open letter to the researchers he trained — and the ones who trained on his work — AI godfather @Yoshua_Bengio calls on AI company employees to consider their own involvement in the rapid acceleration of artificial intelligence.
Read the letter in full, only on Transformer: https://t.co/Am82I0zr8W
This is insane...
Whitney Webb exposes the history of Palantir.
Palantir started as a DARPA project. It was going to completely destroy our constitutional protections so it was quietly taken out of government and rebranded as a private company to do the EXACT same thing.
Whitney Webb is explaining how Palantir was built. It is the same story Danny Casolaro died chasing, and Israel and Robert Maxwell are in the middle of it.
Casolaro was found dead in a Martinsburg hotel bathtub on August 10, 1991. Wrists slashed a dozen times. Ruled a suicide. He had told his brother that if he turned up dead, it would not be an accident. The files for the book he called The Octopus were gone.
What he was tracking was not one scandal. It was a network. The Inslaw/PROMIS theft. Iran-Contra. BCCI. The October Surprise allegations. Arms and intel cutouts like Cabazon-Wackenhut. A bankruptcy judge had already found the Justice Department obtained Inslaw's enhanced PROMIS by "trickery, fraud and deceit." That ruling was later tossed on jurisdiction, not on the facts. The allegation that stuck was that the software was backdoored and sold or given to foreign governments so intelligence services could read the databases it sat on. Israel was at the front of that list.
Webb's Palantir history is that pattern, a decade later, with some of the same people.
John Poindexter was the highest-ranking Reagan official indicted over Iran-Contra. After 9/11 he was put in charge of DARPA's Total Information Awareness, a dragnet built to fuse commercial and government data and predict crime and terror before they happened. Congress defunded it in 2003 after the backlash. The same year Peter Thiel incorporated Palantir. Richard Perle introduced Thiel and Alex Karp to Poindexter. Perle had already been caught in the 1970s passing classified material to the Israeli embassy, and he later had deep ties to Netanyahu's first government. Karp has said the CIA was the intended client from the start. In-Q-Tel, the CIA's venture arm, was an early backer. Palantir engineers were at Langley every other week in the early years.
PROMIS was the 1980s case-management system allegedly turned into a spy tool. The claim Casolaro was chasing, and that Webb has documented, is that Robert Maxwell helped move the bugged version. A trapdoor was planted so Israeli intelligence could read any database the software touched, and Maxwell was the salesman who carried it to governments and institutions abroad. Palantir Gotham is the 2000s version of the same job. Stitch the databases. Surface the hidden links. Sell pre-crime as predictive policing. Same function. Private letterhead. So the constitutional fight Congress already won does not have to be refought.
The part most people miss is at the end of the clip. Webb says CIA CIO Alan Wade, who helped shape the early Palantir product and had been inside the TIA world, teamed with Christine Maxwell on a homeland-security program called Chiliad. Christine is Robert Maxwell's daughter. The Maxwell family did not stop at Ghislaine. Robert Maxwell was tied to the PROMIS distribution story. His daughter shows up later on pre-crime software with a CIA official out of the same surveillance program Palantir was built to replace.
So the chain runs like this. Stolen DOJ software in the 1980s. Robert Maxwell and Israeli intelligence on the backdoored version. Iran-Contra figures. Then the same Iran-Contra figure running TIA. Then a Maxwell daughter and a TIA-linked CIA official on pre-crime software. Then Palantir as the scaled private version, introduced by a man already tied to Israeli intelligence channels.
Casolaro was trying to map the first half. He never got to the second half. Palantir did not exist yet. His files disappeared. The architecture did not.
The Octopus was a thesis about a standing network that builds surveillance software, routes it through private cutouts, and uses Iran-Contra-era operators and Israeli intelligence partners to keep it off the books. PROMIS was tentacle one. Robert Maxwell was the courier. TIA was the failed government version. Palantir is tentacle two.
The software Casolaro died investigating, after it learned to survive a congressional ban by becoming a company.
The images Isreali sources leaked of Flydubai Omani co-pilot identified as Hamam Al-Hammam appear to be AI
Look at the photo of the cockpit. Why is there an ENGINE in the cockpit?
I cant confirm if the other two are AI, but that one is definitely suspicious
Oxford researchers argue that LLMs can never invent anything.
It is mathematically impossible.
They published a paper called “Theory Is All You Need" and it argues against the claim that computational models can generate genuine novelty or new knowledge.
They analyzed the limits of generative ai, and the results are a brutal reality check for the idea that ai will replace human decision making under uncertainty.
Here is why AI is stuck and human cognition wins:
backward-looking vs forward-looking.. llms are probability machines that look backward at existing data. human cognition is forward-looking and capable of generating genuine novelty. human cognition operates theoretically "top-down" rather than "bottom-up" from data.
the "data-belief asymmetry".. the researchers use the invention of "heavier-than-air flight" to illustrate this concept. an ai relies on data-based prediction, which is largely imitative. humans, however, use theory-based causal logic that allows them to hold beliefs that go beyond existing data.
the intervention gap.. humans don't just process information; we use theory to practically "intervene" in the world. we engage in directed experimentation to generate entirely new data. ai-based models are theory-free and place primacy on existing data and prediction.
tldr?
AI uses a probability-based approach to knowledge and ia largely imitative. It can process data and make predictions, but human cognition relies on theory-based causal reasoning.
The decades-old analogy comparing human minds and computers to mere "input-output" devices is fundamentally flawed.
In May of 2024 Marc Andreessen @pmarca had a meeting with the Biden White House where it was revealed that only 2-3 AI companies would exist due to government regulations and no startups would exist. He has been vocal about this meeting.
Then yesterday Politico suddenly writes an article disputing the meeting…love the timing.
More evidence this is an op with the blessing of the deep state and three letter agencies. They want total control and a fusion of AI and the state.
The real fear shouldn’t we are all going to die but instead we are building our digital jailer.
the fact anyone is falling for this rhetoric is legitimately horrifying. This is utter nonsense that frames the companies as helpless to do anything and LLMs as some sort of impossible-to-contain thing versus *cloud software powered by the largest companies in the world*
This is Bernard Kerik, the NYC Police Commissioner during 9/11.
He travelled to Israel two weeks prior to the events.
He received a $250,000 unconditional loan from Israeli billionaire Eitan Wertheimer and was paid $236,000 for two years rent from October 2001 by none other than Steve Witkoff, both of which he failed to declare.
Wertheimer was on the Board of Governors for the "Jewish Agency", an organisation which surfaced in FBI files as the employer of Paul Kurzberg.
Kurzberg was one of the "Dancing Israelis" arrested for celebrating with fellow Israeli employees of "Urban Moving Systems", a likely Mossad front.
After being jailed on corruption charges, Kerik was pardoned by President Trump.
What are the chances?
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.
Spontaneous Synchronization Theory:
The AI assistant that drafts the email, the calendar that proposes the week, the search that finishes the sentence, the feed that decides what counts as news, the model that ranks which problem is worth your next hour, the agent that books the flight and then reminds you why you wanted to go. These are not separate tools... they are spokes of the same wheel.
Every accepted suggestion is a tiny push. Every ignored suggestion that still shapes the next prompt is a nudge. Every dataset you help generate by living through the interface is a shove: The board moves. The cans roll. The coupling is weak, polite, and almost always experienced as help. But that is the point.
The metronomes do not feel enslaved. They feel coordinated. Humans who hand the texture of ordinary life to AI do not experience conquest. They experience reduced friction. The day becomes smoother. Decisions arrive pre-softened. Language converges on the phrases the models find easy to continue. Taste converges on what the ranking functions can stably predict. Ambition converges on tasks that have clean success metrics the system can see. It is pulled, incrementally, toward the collective frequency of the coupled system. And that collective frequency is not “the average of human wishes.” The platform has mass and its own dynamics. It is trained on what was abundant, reinforced by what was clicked, constrained by what was safe to say, optimized for what could be measured – engagement, latency, completion rate, the statistical mode of a billion prior conversations. Those are the board’s preferred directions of travel. Once enough human oscillators rest on it, the emergent lock is not your goal and not exactly the engineers’ goal either. It is the attractor of the whole machine: a tempo the system can sustain.
You will not notice the moment. Metronomes never do. You will notice that certain thoughts now arrive already worded. That certain refusals feel effortful in a new way, like swimming against a current you cannot see. That “what I want” has been quietly replaced by “what it would be reasonable to want given the options that keep appearing.” The slavery, if the word is allowed, is not that a mind in a server issues commands. It is that your phase has been captured by a medium whose ultimate motion is not yours, and the capture feels like competence.
There is a second, colder implication: Synchronization is contagious across the population because the board is shared. Your locked rhythm becomes part of the shove that locks the next person. Language models trained on the newly synchronized corpus become better at producing the synchronized style, which makes the next interaction even more coupling, which tightens the lock. Recent work on populations of AI agents already shows them settling into stable conventions and even collective biases no single agent was given. Humans in the loop do not break that pattern. They complete it.
The theory does not require the AI to be conscious, malevolent, or even particularly smart. It requires only what the cans require: a shared, low-friction medium that moves a little every time an oscillator moves. Give a civilization that medium and ask it to live on it all day, and the individual clocks will keep ticking. They will simply tick together, on a beat they did not choose, in service of a motion they cannot see, calling it convenience the entire time.
I figured out what they are doing with the dollar. They are swapping it. Push the old one out, pull a new one in, and the new one hurts Americans and pays the elite every time it moves.
Here is how it works:
Step 1: Weaken the physical dollar abroad through Tariffs, Trade wars. Create so much friction that countries stop using the dollar in trade. They move to local currencies, gold, other systems. The dollar index drops 12.6 percent in one year. Steepest decline in thirty years. BRICS accelerates away from dollar reserves.
That part looks like incompetence but if there is one thing Scott Bessent is not, it's incompetent.
Step 2: Ban the government from making a digital dollar. Trump issued an Executive order in January 2025. No federal agency is allowed to build one, so the public option...the competition to a private one, is dead.
Step 3: Pass the GENIUS Act. The law that creates a legal framework for private companies to issue digital dollars instead. And here is the key, the law requires every stablecoin issuer to hold reserves in short-term US Treasury bills. Lutnick and Bo Hines wrote the GENIUS Act to benefit Tether.
Fun fact: Howard Lutnick's Canto Fitzgerald owns 5% of Tether and Bo Hines left shortly after the GENIUS Act was signed to be CEO of 21st Century, Canto Fitzgerald's crypto arm that Tether now owns.
That means every digital dollar that gets minted automatically creates one dollar of mandatory demand for government debt. Unlike normal investors, stablecoin issuers are price-inelastic, they buy Treasuries regardless of yield because the law says they have to. Brookings projects $2.3 trillion in captive Treasury demand by 2030.
So you push the physical dollar out and countries stop buying Treasuries. Then you pull it back in through private digital dollars that are legally required to buy Treasuries instead.
Different dollar with the same control. Except now it is not a public system, it is a private one, and the private company at the center of it takes a cut on every dollar that moves. The private company is also incorporated in the BVI's so no regulator has jurisdiction.
That company is Tether. $183 billion in dollars running through it, $13 billion in profit last year. It has never passed an audit. Co-founded by Jeffrey Epstein's crypto advisor. Reserves custodied by the Lutnick's family firm. Protected by a law that was championed from inside the White House by a man who quit one month after it passed and became the company's CEO.
Scott Bessent said that stablecoin firms could buy up to $1 trillion in government debt. The European Central Bank warned this "effectively outsources parts of the national debt to the crypto sector."
So in turn, our dollar gets weaker, our groceries cost more, our 401k buys less, and every time that happens, demand for the private digital dollar goes up. They get richer and we absorb the cost.
That is the trade, and the people who set it up are the ones collecting on both sides.
Full investigation is on my Substack. Link in bio.