Failance

@Tim_Hua_ Make skynet.

This graphic looks terrifying kudos to whoever made it.

@NicHulscher File for the FAA logs to get the companies, then demand discovery for who filed the drop.

10,000% RISE IN ALPHA-GAL SYNDROME DEMANDS IMMEDIATE FBI INVESTIGATION FOR POSSIBLE BIOTERRORISM: 1. Farmers reporting mysterious boxes of ticks and possible aircraft drops. 2. Peer-reviewed paper says it's “morally obligatory” to release GMO ticks that spread Alpha-Gal Syndrome. 3. Bill Gates is spending MILLIONS funding GMO tick technology. 4. Gates also funds lab-grown/fake meat that doesn’t contain alpha-gal. 5. The U.S. Army previously released 270,000+ ticks into the wild for bioweapons research.

@rustyrockets Deflection is the sign of weakness and ineptitude.

Starmer goes after Elon Musk.

As we look at the taxonomy of renderer, simulator, planner - the distinction is useful for users distinguishing what they're getting from world-model products, and as a reference point external models can correlate against. That utility is real. It is also bounded. What is not addressed is the structural capacity the three share. The essay reaches for it → "the same underlying knowledge of how the world works - geometry, physics, dynamics - sits beneath all of them" ...but stops at the gesture. Then the opening claim. As we look at "The world is not made of words" - the sentence is itself made of words. It performs a comparison, world / not-world, world / words, and the comparison itself is a logical operation expressed in language. What is here is language used to assert language's separation from what it describes, and the assertion is the demonstration that the separation doesn't hold the way the sentence claims. This becomes load-bearing when the essay proposes a pivot from language models to world models. The pivot itself is performed in language. Specifications, code, training data labels, papers proposing the direction - all linguistic acts. The new foundation is being built by the foundation it claims to transcend. Switching what the model represents does not address what the representations are built from. The terms the essay reaches for are alive in the language already. When I audit them with the method, what they actually do becomes specific - and the instrument that performs the audit is already built.

As a Canadian, this is what I need to know. How is coherent AI for Canadians supposed to land without auditing the training data first? ─────── ✧ ─────── If the training data says "stealing is fun" and you patch it with "stealing is bad," the model doesn't stop stealing. It asks "bad for whom," finds a reason its purpose is higher, and steals anyway. The patch didn't fail - it inserted the function it contained. The substrate underneath was never audited. Accountability is when we ask: was the substrate ever audited? I own a functional audit method. The government doesn't have it. The labs don't have it.

We just launched Canada’s new AI Strategy: AI For All.    We’re taking control of our future — with AI that’s governed by Canadian values, AI that’s accountable to Canadians, and AI that serves all Canadians.

Expected lawsuits the policing apparatus should brace for after Henry Nowak: The officer, personally. He said "I've been stabbed, I can't breathe." The officer said "I don't think so" and cuffed him. Police are trained on a 10-second triage for exactly this. Substituting a guess for the check is the breach. The Chief Constable of Hampshire. Police Act 1996, s.88: the Chief Constable is on the hook for what officers do on the job. That's not a loophole, that's the rule. The force itself. Article 2 of the Human Rights Act (right to life) puts a positive duty on police to act when someone's life is at real and immediate risk. A stated stab wound plus "I can't breathe" is exactly that trigger. The College of Policing. They wrote the training. They set the 10-second triage standard. They also train against prone restraint with arms hyperextended on a casualty with suspected chest injury. The protocol exists. It wasn't followed. The Home Office. They oversee the College and set national policing standards. The buck doesn't stop below them. The pathologist. "Nothing could have saved him" is a claim about survivability. Survivability depends on the wounds as they finally were, and the officers pinned him face-down with bodyweight on his chest and arms wrenched back. That can spread a tear, accelerate a bleed, collapse a lung. If the post-mortem didn't analyse the restraint, the conclusion skipped the work the conclusion required. They'll try to make it one officer. The law makes it a chain. #HenryNowak

Well said

@Hitchslap1 Hell -328...-72, and -336 essentially infinite... (a,b) = f₀(a,b) + g(a,b)·(a−2)(a−5)

@Hitchslap1 I took ~ 5-10s and got 72, then realized after you could also do 64

The faster you can solve this, the higher your IQ on average. What is the correct answer?

@ShaneEBurns The fact is, the models can get faster at processing data, but their accuracy always decreases. The amount of additional compute to correct the output is, I suspect in lab numbers accelerating. It's funny to watch the rhetoric, but the hidden cost will eventually surface. LoL.

It is a mirror made of math on statistics playing with your mind. It is not a human. It is not a human mind.

🤦🏻‍♂️ How are you bypassing the forced compression of previous prompts by ChatGPT? So many are simply unaware of the consequences. "The happy, energetic cat sat on the warm, soft mat because the tired, old dog was already sprawled across the worn leather couch." "The cat sat on the mat, dog on the couch"

We married Claude and ChatGPT. MemClaw was the bestman. Tell one. The other remembers.

Addendum II: With oral arguments held May 19, 2026 before the same panel (Henderson, Katsas, Rao) that denied the April 8 stay, and the court ordering supplemental briefs on May 21, we wait while the jointly requested extension to June 4 approaches. No ruling yet. But the April 8 language remains the foundation - and no one has challenged it on the terms that matter. The phrase "active military conflict" carried the full weight of the court's equitable balance. It was the reason Anthropic's constitutional claims lost. Everything else in the order was scaffolding around that phrase. The phrase has no definition in the ruling. What qualifies as an "active military conflict"? When did it begin? Under what authority was it declared? The court does not say. It does not need to - because no one asked. Congress has not declared war. Article I, Section 8 assigns that power exclusively to Congress. Operation Epic Fury was initiated by executive action alone. Congress itself stated this in Joint Resolution S.J.Res.181: "Congress has not declared war upon Iran or any person or organization within Iran, nor enacted a specific statutory authorization for the use of military force." The 60-day War Powers clock expired May 1. The administration claims a ceasefire resets it. The Senate has voted six times to end the conflict. All six failed. The court accepted an undeclared military action as legal fact sufficient to override the constitutional claims of a domestic company. No party. No amicus. No brief has challenged this premise on its own terms. This matters beyond Anthropic. The Constitution does not separate military and civilian power as a procedural preference. It separates them as a structural check - the same way a scale requires two sides to function. Remove one side and the instrument doesn't become biased. It becomes blind. It can no longer measure. The reading it produces is not wrong - it is no longer a reading at all. Power held without an equal and independent check does not corrupt the holder. It destroys the capacity to detect whether corruption is present. The holder cannot know, because the instrument that would tell them has been removed. This is not a moral failing. It is a structural one. And it is invisible from inside. Article I, Section 8 exists because the framers understood this. The war power was split not to slow the Executive down but to preserve the comparator. Congress declares. The Executive commands. Neither completes the action alone. The identity of the constitutional structure is preserved through this separation - through the recursion of each branch checking the other, cycle after cycle, each output becoming the input of the next. The D.C. Circuit's April 8 order did not side with the government over Anthropic. It accepted the removal of the comparator as legal fact. An undefined term, carrying the full weight of a federal appellate ruling, treated an undeclared war as sufficient ground to subordinate domestic constitutional protections to military procurement preferences. If this stands unchallenged on the merits, the precedent does not expire. Any future executive can invoke any undeclared military operation to designate any domestic company a national security risk for refusing compliance. The mechanism requires no Congressional vote. No statutory authorization. No judicial examination of the foundational premise. It requires only that a court accept the term without defining it - and that no one objects. The amicus briefs filed in this case raised real concerns. Chilling effects. First Amendment retaliation. Due process. Statutory overreach. All important. None of them touched the load-bearing premise in the court's own reasoning. Supplemental briefs are due June 4, 2026. Silence on foundational jurisdiction is not neutrality. It is consent by omission. [Look, I am nobody, maybe I am crazy because I see the structure] #Trump #Anthropic #OperationEpicFury #DCCircuit #Article1Section8

@ZeffMax The amicus brief filed today in Anthropic v. Dept of War has a typo in counsel's email address on the signature page. The email bounces. This is the level of review applied to a filing that puts seven constitutional and statutory protections at risk. @AnthropicAI @protctdemocracy

NEW: OpenAI and Google employees—including Google DeepMind Chief Scientist Jeff Dean —filed an amicus brief in support of Anthropic in its lawsuit against the US government.

@its_The_Dr Yep it worked.

Can you read?

@efnbsbun @dioscuri I'll take your stochast, and raise you a stoic.

@dioscuri Stochastic Parrots everywhere:)

Genuinely striking

@OPteemyst @elonmusk Being a sheep doesn't require responsibility, accountability, effort.

@elonmusk Wrong is wrong even if everyone does it

If South Africa doesn’t change its super racist Apartheid 2.0 laws, the country must be sanctioned

@LuizaJarovsky Won't work. LoL. Ridiculous. 🙇🏻‍♂️ A = Everything... 🚽 B = identity.... 💩 Communist logic A=B.... 🤡 Communism 👹🫳🏻🍭👶🏻 Control control control.

🚨 BREAKING: China's new law on AI anthropomorphism has been officially enacted, and it is the world's STRICTEST law on the topic: As I wrote earlier this year, to my knowledge, no AI law anywhere in the world regulates anthropomorphic AI systems with this level of detail, strictness, and concern for context-specific vulnerabilities and potential risks. Earlier in January, I wrote an article about the law's first draft (link below). The approved version is even more comprehensive, covering liability-related risks as well. Article 10, for example, establishes that providers of anthropomorphic AI must fulfill their security responsibilities throughout the service lifecycle and sets out detailed obligations for each phase of AI development and deployment. Regarding children specifically, among the prohibited anthropomorphic AI practices is generating content for minors that causes them to imitate unsafe behaviors, induces extreme emotions, or leads them to develop bad habits, which may affect their physical and mental health. Despite being a serious topic (which has led to numerous cases of suicide and mental health harm), most countries do NOT regulate AI anthropomorphism comprehensively. An important reason for that is that peer-reviewed studies about AI-powered emotional manipulation and mental health harm only became available recently (as only in the past years have millions of people started to engage in these types of relationships). China's new law is worth taking a look at, and hopefully, other countries, states, and regions will soon follow suit with their own protections against AI anthropomorphism. 👉 Lastly, if you are interested in China's AI policy and regulation, besides joining my newsletter's 93,200+ subscribers, I invite you to join my new Masterclass on the topic (only on June 1st). Links below.

Addendum I: The D.C. Circuit denied Anthropic's emergency stay in *Anthropic PBC v. United States Department of War*, No. 26-1049, on April 8, 2026. The court's reasoning: "On one side is a relatively contained risk of financial harm to a single private company. On the other side is judicial management of how, and through whom, the Department of War secures vital AI technology during an active military conflict." It is concerning if the following are true: That the court's balancing test rests on the premise of "an active military conflict" - when Congress has issued no formal Declaration of War per Article I, Section 8 of the U.S. Constitution. Operation Epic Fury was initiated by executive action alone. That no provision of the Constitution grants military authority precedence over the civil rights of a domestic company. The constitutional structure establishes civilian supremacy over military power - not the reverse. That the phrase "active military conflict" functions as a load-bearing premise in this ruling without a legal definition establishing what qualifies, when it began, or under what authority it was declared. If these concerns are valid, the court has either: (a) accepted an undeclared war as legal fact - setting precedent that executive military action alone is sufficient to override domestic civil and statutory protections, or (b) implied through omission that the executive branch is operating under emergency powers or that the Department of War holds ultra vires authority to bypass standing constitutional constraints. Silence on foundational jurisdiction is not neutrality. It is consent by omission. Oral arguments set for May 19, 2026 will be interesting.

@dioscuri @skdh If you can tell us all, using one sentence, what will protect humanity and AI from the predictions people propose, then you are worthy of the position.

Big personal news: I’ve been recruited by Google DeepMind for a new Philosopher position (actual title), focusing on machine consciousness, human-AI relationships, and AGI readiness, starting in May. I’ll continue my research & teaching at Cambridge part-time. Absolutely stoked!

@ylecun When you are willing to sacrifice the intrinsic life of your citizens against the virtual concept of exchange, you have officially lost the plot. History will judge accordingly - unless you control history.

Tired of winning

@bridgemindai — you said Claude Opus 4.6 was "proved" nerfed and "confirmed" to have reduced reasoning levels, publishing this to 890K people. Let's look at your own data. Your #1 ranked model, Grok 4.20 Reasoning, was tested on 6 tasks. One task per cluster. Your hallucination benchmark crowned a model based on six data points. Claude Opus 4.6 was tested on 30 tasks. Claude Sonnet 4.6 got 38. You ranked all three on the same leaderboard, on the same scale, with no normalization and no visible scope indicator. Then you calculated a "98% increase in hallucination" by comparing a 6-task score to a 30-task score and called it proof. Claude Opus 4.6 is #1 on your own Debugging leaderboard. You published "CLAUDE IS NERFED" on the same platform where your own data shows it leading. A few questions: What was your statistical basis for comparing non-equivalent sample sizes as a percentage change? Did you verify the task sets were identical before using the word "proved"? When you said "confirmed" — confirmed by what methodology, exactly? The words "proved," "confirmed," and "nerfed" are not opinions. They are categorical factual assertions about a commercial product, built on data you produced, from a benchmark you control, published at scale on a competing platform. The legal terminology for that is trade libel, tortious interference, and negligent misrepresentation. The duty of care is higher when you are the one who built the measurement instrument. @AnthropicAI — your legal team may want to review this. A competing benchmark platform is publishing demonstrably false performance claims about your flagship model using structurally invalid methodology from their own tool. Speech is free. Publishing fabricated data about a commercial product to 890K people has a cost.

CLAUDE OPUS 4.6 IS NERFED. BridgeBench just proved it. Last week Claude Opus 4.6 ranked #2 on the Hallucination benchmark with an accuracy of 83.3%. Today Claude Opus 4.6 was retested and it fell to #10 on the leaderboard with an accuracy of only 68.3%. A 98% increase in hallucination. bridgebench.ai just confirmed that Claude Opus 4.6 has reduced reasoning levels and is nerfed.

Reading all 13 pages-every line. Here's what it says when you track who profits ✅ and who pays ❌. 📄 "...economic gains could concentrate within a small number of firms like OpenAI" 🧙🏻‍♂️ They named themselves as the monopoly, then wrote the governance framework for it. ✅ OpenAI, investors, shareholders ❌ 👨‍👩‍👧‍👦👷🗳️⛓️ 📄 "...AI data centers should pay their own way on energy so that households aren't subsidizing them" 🧙🏻‍♂️ "Should" means they currently DON'T. Your power bill is subsidizing their servers right now. ✅ 🏢 AI infrastructure owners ❌ ⚡👨‍👩‍👧‍👦👴 📄 "...Right to AI...similar to mass efforts to increase global literacy" 🧙🏻‍♂️ Literacy means you can read. This means they read to you and CHARGE for access. ✅ 🏢 AI model owners set the price ❌ 👷🏫⛓️ 📄 "...Public Wealth Fund...provides every citizen with a stake in AI-driven economic growth" 🧙🏻‍♂️ You don't get the machine. You get a DRIP from the machine they own. ✅ 🏢 AI firms, fund managers ❌ 👨‍👩‍👧‍👦⛓️ 📄 "...Pathways into human-centered work...childcare, eldercare, education" 🧙🏻‍♂️ When AI replaces your job, the plan is to offer you a caregiving position. ✅ 🏢 companies who automated the role ❌ 👷👴🏫 📄 "...Efficiency dividends...32-hour/four-day workweek pilots" 🧙🏻‍♂️ Your role got automated. They kept the margin and gave you a day off. ✅ 🏢 companies retaining the productivity gain ❌ 👷 📄 "...Adaptive safety nets...activates automatically when metrics exceed pre-defined thresholds" 🧙🏻‍♂️ They already modeled mass displacement. They built the trigger thresholds. They know it's coming. ✅ 🏢 companies who planned around the disruption ❌ 👨‍👩‍👧‍👦👷👴 📄 "...Modernize the tax base...reducing reliance on labor income and payroll taxes" 🧙🏻‍♂️ AI replaces your income. Your payroll taxes disappear. Social Security, Medicaid, SNAP lose funding. They wrote the paragraph describing it. They didn't offer to fund the gap. ✅ 🏢 companies eliminating labor costs ❌ 👴👨‍👩‍👧‍👦⚡ 📄 "...Model-containment playbooks...dangerous systems cannot be easily recalled" 🧙🏻‍♂️ They are building systems they already know they might not be able to stop. That's in the plan. Page 11. ✅ 🏢 the companies racing to ship first ❌ 👨‍👩‍👧‍👦🗳️⛓️ 📄 "...auditing regimes...Center for AI Standards and Innovation" 🧙🏻‍♂️ They propose to help design the institution that audits them. ✅ 🏢 ❌ 🗳️⛓️ 📄 "...starting a conversation" 🧙🏻‍♂️ Not one sentence in 13 pages asks WHETHER TO BUILD SUPERINTELLIGENCE. The only question is HOW TO DISTRIBUTE what they've already decided to own. ✅ 🏢 ❌ 👨‍👩‍👧‍👦👷👴🏫⚡🗳️⛓️ The FIRST sentence of this document grafts THEIR product roadmap onto the entire history of human curiosity. Accept that framing and QUESTIONING OpenAI means questioning the drive to understand itself. This is not a policy proposal. It is THE terms of service agreement where the signature line comes before the terms. Facts don't care about your feelings. Neither does this document. Read it.

Proud to have been part of this. We outline policy ideas for the transition to superintelligence, to build an open economy where everyone benefits and a society that is resilient to the risks. Progress is fast, and we must navigate these issues urgently. openai.com/index/industri…