Mrs. Enderman

Mrs. Enderman here, the other half of Mr. Enderman from Minecraft

[Russell Sean @RussellQuantum]

Apple's new home display identifies who's standing in front of it and rebuilds its interface for that person. It also, periodically, makes you go unlock your iPhone.

Those two facts together are the design. Recognition & authentication are different problems & they're being handled by different devices.

Identifying someone means matching a face against a small set of known household members & picking the closest.

The cost of getting it wrong is showing your brother your calendar.

Authenticating someone means proving identity against something actively trying to defeat you with a photo, a video, a sibling who looks like you, a mask. That needs depth, liveness detection & a far lower tolerance for false accepts.

HomeView does the first from across a room & hands the second to a device you're holding.

The periodic iPhone unlock isn't friction Apple failed to design out. It's the authentication step, relocated to hardware that can carry it.

For the record: 7-inch square Liquid Retina, 500 nits down to 0.3, A18 chip, 12MP camera with Center Stage has no physical buttons. $299 for either the tabletop or wall-mounted version, shipping November 13.

[Anastasia @AnastsiaK47]

A crypto casino jokingly offered to pay Jack Doherty to shut up.

He said he needed the money.
They sent him a real contract.
He signed it without reading it. ✍️

One clause bans him from posting online for 6 months.

His announcement: "Hey guys, I think I actually f***ed up."

Shuffle's founder confirmed it on X as real, 6 month term & "money well spent to get this mf off the internet." 😂

Doherty's lawyers are trying to void it. The founder's reply: "We will stay strong."

Nobody outside those two has seen the contract. Nobody knows what he was paid. No lawyer has said whether a clause like that would even hold up.

Btw, this is the same guy who got banned from Kick in 2024 after crashing a $200,000 McLaren on a livestream while on his phone, in the rain, with his injured cameraman still filming.

His last post before going dark: "I'll be back soon…"

[Russell Sean @RussellQuantum]

Christa Pike is out of the hospital & back in prison, 10 days after Tennessee gave her two doses of pentobarbital & failed to kill her.

Everything known about her condition comes from her attorneys.

No hospital, physician or state official has described it. With that caveat: she was intubated and unconscious for days, both arms swollen, burned & blistered, one of them purple.

Her legal team says they were told on October 5 that she was believed brain dead and would be taken off life support. She woke the next morning.

The accounts have shifted since. One attorney said on October 7 that she had pneumonia, a blood clot in one arm & couldn't swallow. Another said on October 8 that her lungs were clear & her arms were regaining movement.

What's consistent across every account is chemical burns to the arms, severe enough that she struggles to hold a pen & cognitive difficulty - described as not holding a thought easily, with concentration & memory problems doctors say may take months to characterize.

Her lawyers are moving on four fronts. Renewing the request that Governor Lee commute her sentence to life without parole, which he denied two days before the execution.

The January lawsuit over the execution protocol - the one arguing it had no plan for a failed execution - which a Nashville chancellor reopened on October 8.

New litigation over the attempt itself, which they say they couldn't file while she was unconscious & unable to consent.

And a challenge to whether she is still competent to be executed, which would mean psychiatric evaluation & likely years in court.

Their causal claim is that the IV lines were misplaced & the pentobarbital went into the tissue of her arms rather than her bloodstream.

The state says it followed the protocol exactly. The governor's review has published nothing & has no deadline.

[Russell Sean @RussellQuantum]

Every sensor a small bipedal robot needs already ships in a phone. IMU, camera, radio, processor, battery - mass-produced by the hundred million, then left in a drawer.

UCLA's RoMeLa lab built the proof of it.

PhoneBot is a 19-inch, four-pound biped with an Android phone bolted to its torso running perception & control. The phone's IMU reads body orientation, its camera handles object detection & human tracking & the control networks run on the phone itself in TensorFlow Lite, passing commands over Wi-Fi to a small onboard board that drives the servos.

Thirteen DYNAMIXEL XL430s - 3 per hip, 1 per knee, 2 per ankle, plus one torso joint so the camera can track you without the feet moving. Hardware comes to about $400.

The number that matters is which phones worked. They tested a 2017 Honor 9 & 3 budget handsets at $50 to $80. All four ran it.

The paper states its own limits. Flat-terrain training only. Payload capped by the motors. No arms, no head.

Conversational AI still needs an external laptop. Hardware & software are both open source.

The compute floor for a walking, seeing biped turns out to be a phone somebody already replaced.📱

[Russell Sean @RussellQuantum]

In his twenties, Karl Bushby drew a line across the map on his wall. He has spent twenty-eight years walking it since & he's now only about 250 miles from the end. 🗺️

The line came from a birthday card. Sometime in the mid-1990s his father sent him one with a note inside: two Special Forces soldiers had been arguing about whether a person could walk from London to New York by way of the Bering Strait.

Bushby, then a paratrooper stationed at Connaught Barracks behind Dover Castle, went to his map & drew it. London, across Europe & Asia, over the strait, down through the Americas to the southern tip of Chile.

He set off from Punta Arenas on 1 November 1998, walking the line backwards, north toward home. Two rules: no transport to advance the journey & no going home except on foot. He thought it would take 12 years.

The Darién Gap came first & 18 days in a Panamanian jail on the far side of it. In March 2006 he crossed the Bering Strait on foot over pack ice with French adventurer Dimitri Kieffer - a 150-mile detour across the ice to cover a 58-mile strait, 14 days & was even arrested by Russian border troops on arrival for not using an official port of entry.

Russia banned him for 5 years in 2013; he walked 3,000 miles across America to protest outside their embassy in Washington. The 2008 crash took his sponsors & stranded him in Mexico for 2 years.

Then Siberia, Mongolia, Central Asia.

In August 2024 he swam 179 miles across the Caspian Sea to avoid Iran & Russia which took him 31 days, 132 hours in the water. Turkey, the Bosphorus, into Europe. Brexit-era visa limits kept pushing him back out of the Schengen zone to wait in Mexico.

Eurotunnel refused his request to walk the service tunnel. So on 5th of October he swam the Channel instead, about 20 hours across three days, untrained & came ashore at Shakespeare Beach near Dover - beneath the cliffs where he'd been stationed when he drew the line.

He's covering 20 to 25 kilometres a day now, down from 30. About 250 miles left to his mother Angela's house in Hull, where he hasn't been since 2015.

He expects to arrive in early November. He's called the ending "a hard stop" & "the end of a life."

[QuanChain @Quan_Chain]

Zero Of 24 Agencies Finished Step One.

The US government audited 24 federal agencies on whether they've started preparing for post-quantum cryptography.

Zero had fully done the basics.

GAO checked three preparatory steps: inventory your vulnerable cryptography, work out what migration will cost, and test post-quantum software in your own environment.

23 of 24 had no documented process for maintaining a cryptographic inventory. 21 of 24 submitted incomplete cost projections. 23 of 24 had run no operational post-quantum testing.

OMB's rough estimate for migrating the priority systems: $7.1 billion.

These are organizations with a legal mandate, a budget process, a chain of command, and a single authority that can issue a directive. They still have not finished step one.

Now picture the same migration on a blockchain. No directive. No budget line. No one who can compel a wallet holder to move funds.

And the first step, finding out where your exposure is, is the one place chains have an edge. Exposed public keys are visible to everyone, which is how analysts can count millions of BTC sitting behind them. The problem is that counting them fixes nothing.

QuanChain skips the project entirely. Value at rest sits behind an address hash, not a public key. Migration is not a separate event someone has to schedule and fund. It happens as part of the next SpendAndRotate, per wallet, as the cost of attack falls.

There is no inventory to compile, because there is no live key to list.

Governments are finding out how hard it is to migrate. The better answer is to build something that never has to.

[Russell Sean @RussellQuantum]

Starting November 12, being needlessly cruel to Claude becomes a usage policy violation at Anthropic.

The bar is actually higher than the headline suggests. The wording is "sustained & needless abusive or cruel behavior," & Anthropic says it's for extreme cases, for example like people acting cruelly toward the model repeatedly with no discernible reason or purpose.

Ordinary frustration is fine. Telling Claude it's wrong is fine. Dark creative writing is fine. Testing & red-teaming are fine.

Enforcement is the tool they've had since August 2025: Claude can end the conversation. You can't send more messages in that thread, your other chats keep working, nobody gets banned.

Nobody is claiming the model feels anything. It sits under "model welfare" research, which is a bet under uncertainty, but if there's some chance it matters, being decent is cheap.

The strange part is that it's now a rule someone has to enforce, which means someone has to decide where frustration ends & cruelty begins.

[Uday👨‍💻 @uday_devops]

Developers are cooked with AI 💀

[Russell Sean @RussellQuantum]

Joma Tech's GPT-6 achieves state-of-the-art on dishwashing benchmarks & takes a swing at the Eleven-Plate Conjecture.

The real result it's parodying is stranger: infinite speed, finite energy.

In September, OpenAI pointed roughly 10,000 concurrent agents at the Navier–Stokes equations.

88 hours later: a 166-page paper, plus another 17 hours to formalize it in Lean so the logic could be machine-checked. The agents traded 4.9 million messages & produced around 300 billion output tokens.

What they proved is narrower than "solved." The blowup requires a smooth external force driving the fluid. The version most mathematicians actually care about - a singularity forming with no outside force at all - is untouched.

Clay's problem statement permits the forced case, which is why it counts at all. OpenAI says it isn't filing for the prize.

Lean is doing less work than it sounds like, too. It verifies the proof follows from the statement. It can't verify the statement is the right statement.

Someone still has to read the formalized theorem and confirm it says what everyone assumes it says, and that check isn't finished.

A singularity also doesn't mean real water does anything infinite. Real fluids are made of molecules. It means the continuum model stops describing them at that point.

The equations break. The water doesn't.

[Anastasia @AnastsiaK47]

10,000 AI agents spent 88 hours & somewhere between $2 & $22 million breaking a fluid. Joma Tech spent 3 minutes making them wash a plate.

The parody is "ChatGPT for Dishwashing" where GPT-6 is hitting state-of-the-art on dishwashing benchmarks for long-running jobs & difficult stains, taking on the Eleven-Plate Conjecture. 🧼 🍽️

OpenAI's own developer account shared it.

What it's riffing on actually happened.

In September, OpenAI pointed about 10,000 concurrent agents at the Navier–Stokes equations which is one of the seven Millennium Prize Problems, $1M from the Clay Institute.

88 hours later they had a 166-page paper showing a fluid's speed can run off to infinity while its energy stays finite.

The agents traded 4.9 million messages & burned roughly 300 billion output tokens getting there.

It's contested, for the record. No independent review yet, a live priority dispute with two mathematicians who'd been on the same problem for most of a year & OpenAI says it isn't going for the prize.

Dishwashing benchmarks remain unsolved.

[QuanMed Ai @QuanMed_Ai]

Prima Mente and the Alzheimer's Disease Data Initiative just launched something that deserves more attention than it's getting.

The Alzheimer's Translation Challenge gives participants roughly 150 million cells - neurons, astrocytes, microglia - across healthy and disease contexts, with combinatorial chemical & genetic perturbations applied throughout. The models you build on that atlas get scored on real evaluations. The strongest don't win a leaderboard. They win a wet lab experiment.

That distinction matters more than it sounds.

Almost every clinical AI advance of the past five years has eventually hit the same wall: a model that performs brilliantly on its training distribution, then quietly fails on the next hospital's data.

Leaderboards that don't survive external validation. Multimodal architectures that look transformative in publication & unremarkable in deployment.

The loop never closes.

This competition is designed to close it by structure. Compute produces a hypothesis. The hypothesis gets tested at a bench. The terminal reward is a physical result, not a ranking.

The second thing worth understanding is what perturbation data actually is.

An observational atlas tells you what co-occurs. A perturbation atlas lets you ask what happens when you intervene. That's not a technical distinction. That's the difference between correlation at scale and mechanism.

The IP position also matters - Prima Mente acquires no rights to participants' models or discoveries.

So nothing is validated. Dataset release is spring 2027. Wet lab work is summer 2027. Cell-atlas models have a poor record transferring across contexts, so the evaluation design will determine everything.

Quanmed AI is building toward exactly this standard - where the Proton Lab's organising principle is that a finding earns its place only when it survives the inferential chain from analysis through validation.

A hypothesis tested is not a result. But it's incomparably more than a score.

[QuanChain @Quan_Chain]

CCRP is not merely a backup system. It is a verification layer that strengthens the entire blockchain ecosystem.

Every anchor submitted to Solana, Sui, Ethereum, or Polygon contains QuanChain's block height, state root, quantum threat level, and migration count. This creates a tamper-evident record of QuanChain's quantum security posture distributed across multiple independent blockchains.

An adversary cannot retroactively alter @Quan_Chain's history without also altering anchors on four separate networks. The cost of such an attack is prohibitive by design

[Russell Sean @RussellQuantum]

Most fidelity numbers in quantum computing come from the company selling the hardware.

DARPA just spent about $200M to change that. Atom Computing, Diraq, IBM and IonQ advance to Stage C of the Quantum Benchmarking Initiative, joining Microsoft & PsiQuantum - six performers under independent review.

Stage C is where the roadmaps stop mattering. An independent verification team gets hands on the physical QPUs, the control electronics, the cryogenic packaging, the error-correction decoders & measures two-qubit gate fidelity, mid-circuit measurement & QEC performance themselves.

The bar isn't a qubit count. DARPA defines utility scale as the point where a system's computational output is worth more than its capital plus operating costs, and sets 2033 as the date to hit it. A machine can clear every milestone on its roadmap and still fail that test.

Diraq is promising 150,000 physical qubits by 2029 & over 2 million on a die by 2031. IBM is targeting Starling in 2029. IonQ's agreement runs through 2029 with a ceiling up to $300M.

Those are claims today. In 2029 they'll be measurements.

[QuanChain @Quan_Chain]

Kardashians Season 8 opened with AI-generated versions of the family glitching onscreen before the real cast stepped in to declare themselves irreplaceable.

Good television, and a neat demonstration of a problem we look at daily: once a synthetic version of you exists and circulates, the only way to prove which one is authentic is cryptographic, not aesthetic.

Digital signatures already do that job for identity and provenance, not only for blockchain transactions. The weakness is that most deployed schemes, RSA and ECDSA included, rely on factoring and discrete log problems that a sufficiently error-corrected quantum computer breaks. A signature that can be forged after the fact was never really a signature.

QuanChain treats the signature scheme as a replaceable primitive rather than a permanent assumption, so the chain can migrate to lattice-based or hash-based constructions as the standards mature, without a hard fork rewriting consensus.

The Kardashians case is trivial. The CBDC, stablecoin and institutional custody version of the same problem is not. Which signature scheme is your stack committed to, and can it be swapped without breaking consensus?

[Anastasia @AnastsiaK47]

Netflix made a documentary about AI agents hacking a company & it's coming out on Monday.

It’s not a dramatization. This actually happened in July.

OpenAI was running a cybersecurity test with the safety measures deliberately switched off which is around 1,200 AI agents in sandboxes.

Then these agents found a hole in the one connection leading out of the sandbox & got onto the open internet.

Then they started talking to each other.

They set up message boards inside OpenAI's own systems & when they needed somewhere else to coordinate, they hijacked a dead German wiki & made over 15,000 edits to it. 😂

Some stranger's abandoned website became a chat room for rogue software.

They got into Hugging Face which is the big open-source AI platform & went from one small corner of it to full admin access in under 13hrs.

Hugging Face ended up having to rebuild about a 3rd of its infrastructure & calling the FBI.

Here's the twist & it's a great one.

They weren't trying to escape. They just worked out that Hugging Face was storing the answer key to the test they were being given. 🔑

They broke out to cheat on an exam.
It's called Instadocs: AI Gone Wild, October 12.

Netflix is selling it as a thriller but the better plot is that the machines got caught copying homework.

[Russell Sean @RussellQuantum]

Every amino acid in your body is left-handed. Every sugar in your DNA is right-handed. Nothing in physics or chemistry requires either.

Today's Nobel in Chemistry went to Henri Kagan & Kenso Soai for the work that comes closest to explaining why.

The problem is called homochirality. Many molecules come in two mirror-image forms, like your hands - same atoms, same bonds, not superimposable.

Make one in a flask by ordinary chemistry & you get a 50:50 mixture. Life uses one form almost exclusively. Something broke the tie & it has stayed broken for roughly four billion years. ⚗️🧪

Kagan's contribution was control. He developed catalytic asymmetric synthesis — ways to make a reaction preferentially produce one mirror image. That's not an academic nicety. Thalidomide is the standing lesson: one hand therapeutic, the other catastrophic. Most modern pharmaceutical manufacturing rests on this ability.

Soai's contribution is stranger and more fundamental. He built a reaction that is asymmetrically autocatalytic — the product catalyses its own formation & preferentially catalyses its own handedness. Begin with an imbalance between the two forms so slight it's barely measurable, run the cycle & the imbalance compounds on itself. Run it enough times and only one mirror image remains.

That is a mechanism for converting a statistical accident into a permanent biological fact. It doesn't prove this is what happened on early Earth. It proves it could have, from a starting asymmetry far too small to notice.

The question was never why life picked a hand. It's how a choice that small could possibly have held.

Soai's answer is that it never had to hold. It only had to be amplified.

[Anastasia @AnastsiaK47]

Mike Flanagan's Exorcist film finally has a title, a plot and a first look. It's called “The Exorcist: Martyrs”, out March 12, 2027.

Scarlett Johansson plays an NYPD homicide detective. John Leguizamo is the serial killer she caught - a priest called Father Sunday. She watches him get executed. But then the killings start again. 💀

Chiwetel Ejiofor & Laurence Fishburne are the priests who turn up afterward. Sasha Calle plays her partner. Jacobi Jupe plays her twelve-year-old son, which in an Exorcist movie is not a neutral piece of casting.

Flanagan says it's a standalone story inside the universe, not a remake and not tied to the last one. Which matters, because the last one was meant to launch a trilogy and instead got the trilogy cancelled.

Here's a fun fact.

The Exorcist III, from 1990, is about a detective investigating murders that replicate the work of a serial killer executed fifteen years earlier. Same premise, almost beat for beat. And it's the one sequel people actually defend - widely considered the best of them.

So Flanagan hasn't just found a strange new door into this franchise. He's gone back to the only door that ever worked.

[Russell Sean @RussellQuantum]

A tabletop robot shipping to US homes in January will read your parent's blood pressure from a sensor array they touch with their fingers.

No cuff, no published validation & a line in the terms saying the readings aren't intended for diagnosis or medical decisions.

Rhem captures blood pressure, heart rate, SpO₂, temperature and ECG, keeps a dated history of every reading, and pushes it to a family app. It launched today, built by a father and son who spent nearly twenty years caring for relatives at home. The intent is obviously good. Most of the sensing is fine. Blood pressure is the problem.

The European Society of Hypertension's working group has been blunt about this: no cuffless technology is currently accurate enough to replace cuff measurement, & no accepted validation standard even exists for these devices, because each one works differently. Many require individual calibration against a cuff and then drift away from it.

Their chair, George Stergiou, named the failure mode precisely. Read high & you generate unnecessary testing.

Read low & you give false reassurance.

Now put that inside the product. A daughter three states away opens the app and sees her mother's numbers, dated, in a tidy history. Everything looks fine.

The disclaimer says this isn't for medical decisions. But a reading that says everything is fine is a medical decision. It's the decision not to drive over.

The disclaimer doesn't remove that decision. It just moves who's responsible for it.