Russell Sean (@RussellQuantum) header
Russell Sean (@RussellQuantum) profile picture

@QuanMed_AI- medical robotics research system by AI and Quantum not big Pharma on @Quan_Chain the only auto-migrating quantum chain. MSc Neuro/AI-Robotics Dev

Exeter
Joined September 2022
Photo by Russell Sean (@RussellQuantum)Photo by Russell Sean (@RussellQuantum)Photo by Russell Sean (@RussellQuantum)Photo by Russell Sean (@RussellQuantum)Photo by Russell Sean (@RussellQuantum)Photo by Russell Sean (@RussellQuantum)Photo by Russell Sean (@RussellQuantum)Photo by Russell Sean (@RussellQuantum)Photo by Russell Sean (@RussellQuantum)Photo by Russell Sean (@RussellQuantum)
Google put a quantum chip on the cover of Science this week for beating a bound of 8/9. That sounds small. It's one of the cleanest demonstrations in physics. The test is the Mermin-Peres magic square. Fill a 3×3 grid so every row multiplies to +1 and every column multiplies to −1. With fixed numbers that's impossible, and you can see why in one line: multiply all the rows and you get +1, multiply all the columns and you get −1, but both products contain the same nine entries. The best any strategy using predetermined values can manage is 8 of the 9 constraints. 88.89%. Willow hit 98.30% ± 0.01% over 900,000 trials, on four qubits. The reason is contextuality. Quantum observables don't carry fixed values waiting to be read off — what you get depends on which compatible measurements you make alongside it. There's no consistent grid to fill because the entries aren't numbers sitting there in advance. They also ran a Kochen-Specker-Bell test at χ = 5.618 ± 0.005 against a classical bound of 4.0, a 71-qubit GHZ parity game that beat the classical rate at every size with genuine entanglement verified at 45 qubits, and a 2D hidden linear function test on constant-depth circuits across 105 qubits. Now the part the headlines are mangling. This is not a computational advantage result. The paper claims measurable separations in bounded-resource tasks, which is narrower. The test circuits are Clifford-based and stay efficiently simulable classically — you can run them in Stim on a laptop. Beating a game's classical bound is not outrunning classical hardware. Where it's genuinely useful is diagnosis. These tasks demand exact multi-qubit phase agreement, which makes them brutal on small coherent over-rotations — precisely the errors random-circuit sampling averages away. That's a calibration reference that doesn't flatter the machine. Pichai framed contextuality as possibly the deeper reason quantum computers outperform classical ones. That's a real research position, not a marketing line — contextuality is the known resource behind magic states in fault-tolerant schemes. A result you can simulate classically is still a result. It's just a different one than the headline suggests.
4
8
388
Pouring seawater on top of sea ice makes more ice. That sounds like it shouldn't work & the reason it does is the cool part. 🧊 A trial at Cambridge Bay, Nunavut, over winter 2024–25: eight test plots, three controls, flooded once or twice with up to 20 cm of seawater. By the end of winter the treated plots were up to 32 cm thicker than the controls. Twice-flooded beat once-flooded. Through the melt season the treated ice stayed brighter and melted more slowly. Two things are happening here. The obvious one: pumped seawater sitting on a surface far below freezing turns into ice. So now you've added ice. The other matters more. Sea ice grows downward, from the underside, as heat conducts up through the ice and escapes into the air. Snow lying on top is an excellent insulator - it blocks that heat loss and throttles the growth. Flooding saturates the snow and destroys its insulating value. The ice underneath then starts growing faster on its own. So now you aren't just adding ice. You're also removing the blanket. Now the other half. A 2016 study put the requirement for covering 10% of the Arctic Ocean at roughly 10 million wind-powered pumps. The whole ocean, 100 million. A 2025 review concluded the approach is "simply not feasible for use at a scale and at a rate that would be meaningful for sea ice protection." And the Cambridge Bay work was co-authored by the person running the company that sells the technique. Normal for early field trials. Still worth saying. The physics isn't in dispute. Eight plots got thicker, for reasons that make sense. The question was never whether you can thicken a patch of ice. It's whether you can do it across millions of square kilometres, every winter, indefinitely, while the thing melting it carries on. ❄️
2
2
8
545
A breach of 16,000 records sounds trivial at first. Equifax was 147 million. This one may be worth more. n0n published emails, phone numbers and account IDs for roughly 16,000 Chase customers. The data appears to come from the Reserve program which is an invitation-only card reportedly requiring around $10 million in assets under management. Record counts are a poor measure of a dataset. What matters is how far the data narrows the search space. A random sample of 16,000 American adults tells an attacker nothing about who is worth calling. Expected yield is the base rate, and the base rate for anyone holding eight figures at a single institution is very low. Condition on $10 million in assets under management and that base rate goes to one. Every name on the list has already cleared the threshold. The screening is finished. Chase performed that screening. They verified the assets, ran the underwriting, issued the card. The attacker inherited it at zero cost. This is why breach severity scales so badly with record count. The information isn't in the rows. It's in the selection rule that produced them. The entry point, per Chase, was third-party vendors holding marketing data. The bank's own systems weren't touched. None of it independently verified yet. But a filter works the same whether or not you were the one who built it.
4
6
292
Anthropic priced its IPO at $83 a share earlier today, the largest offering on record. Hours earlier, a coalition that agrees on almost nothing published a statement saying the company is not a safe AI company. The offering: 980 million Class A shares, roughly $81.3 billion gross, a reported $1.5 trillion fully diluted valuation. It was upsized hard — launched at $70 to $78 for 825 million shares, landed at $83 for 980 million, with the book reportedly six to 7 times oversubscribed. Revenue in the S-1 is $33.2 billion through nine months, Q3 around $17 billion, up about twelvefold year over year. Trading opens Thursday under ANT. The statement is called MisAnthropic, organized by the Center for AI Safety. Its argument is that Anthropic warns publicly that recursively self-improving AI could concentrate power and escape human control, then pitches that same capability to public-market investors as a competitive advantage. It cites a February revision of the company's Responsible Scaling Policy that dropped unilateral commitments to pause & lobbying against SB 1047, the RAISE Act & the AEGIS Act alongside private advocacy for federal preemption of state AI law. What makes it unusual isn't the argument. It's the signature page. Steven Pinker. Judea Pearl. Oliver Hart. Laurence Tribe. Lawrence Lessig. Three Fields Medalists. Max Tegmark, who thinks the risk is existential, signed alongside Guillaume Verdon, who built an accelerationist movement around the opposite position. So did Hugging Face's Thomas Wolf & xAI co-founder Igor Babuschkin. Those people agree on nearly nothing. They agree on this.
4
6
541
Tesla just fit a transformer, an inverter & a breaker into one box. It's called Powercore, unveiled by Tesla Megapack this week: a solid-state medium-voltage transformer that does the job of three separate pieces of grid hardware. 7.4 MW of continuous power at 34.5 kVac. 98.86% peak efficiency. AC and DC across a wide voltage range, all in one modular enclosure. No flammable oil inside. It's aimed at one of the grid's slowest chokepoints: transformer lead times. Tesla is pitching it for faster solar installs, off-grid microgrids paired with Megapack batteries & direct 800 Vdc power for data centers. One unit can serve 2,000 to 3,000 homes. Musk says it's been in development for more than 3 years. Production starts in 2027 & NVIDIA is a partner for the data center side.
1
4
13
518
Isomorphic Labs raised $5 billion today, making it the largest venture round in biotech history. The company has never put a drug into a human. That's not a knock. It's more like a statement about what's being priced. Isomorphic spun out of DeepMind in 2021 to build drug design on top of AlphaFold, the protein structure prediction system that earned Demis Hassabis & John Jumper a share of the 2024 Nobel in chemistry. The round was co-led by Greenoaks & Sequoia, with a16z new & Alphabet continuing. 6 months ago they raised $2.1 billion. External capital is now past $7 billion. The pipeline is oncology & immunology & the company reports "positive preclinical results." No data published. No candidate dosed. Roughly 90% of drugs entering Phase I never reach approval & the 2 dominant causes are toxicity & lack of efficacy. Lack of efficacy usually means the biological hypothesis was wrong - the target you chose doesn't actually drive the disease. Structure prediction addresses neither. AlphaFold tells you the shape of a protein to remarkable precision. It doesn't tell you whether binding that protein helps a patient & it doesn't tell you what else your molecule touches on the way through a body. None of which says Isomorphic fails. Better molecules against a correct target are genuinely valuable, and faster iteration on binding affinity is real progress. It's a reason to be precise about which part of the attrition curve this technology is pointed at & it isn't the part where most drugs die. The 1st readout that counts is a Phase I dose. Until then there's nothing separating a company that has solved something from one that has raised money.
2
3
1
24
1,640
100 years ago Robert Goddard stood in a field in Massachusetts next to a frame of pipes & launched a rocket 41 feet. On Saturday a tower in Texas caught one coming back from orbit. SpaceX has been catching boosters since 2024. This was the 1st catch of an upper stage & the gap between those two dates is a physics problem rather than a scheduling one. The booster separates at roughly 1.5 km/s & never properly leaves the atmosphere. The ship comes back from orbit at about 7.7 km/s. Kinetic energy goes as v², so half v-squared puts the booster near 1.1 megajoules per kilogram to shed & the ship near 30 which is more than 20 times the energy per kilogram of vehicle. Essentially all of it has to go into the atmosphere as heat, because you cannot carry enough propellant to brake from orbital velocity. That's why the booster is mostly bare steel & the ship needs a heat shield. Flight 15 flew new tile-retention hardware, a curved tile design & fixes for the gaps where plasma could work in behind the tiles. Three of the 26 Starlink satellites it deployed carried cameras pointed back at that shield. Then, having survived entry, the vehicle has to arrive above a specific tower with enough propellant left for a flip and a three-engine landing burn. The booster carries margin. The ship arrives nearly empty. It worked. Ship 42 was in the tower's arms 22.5 hours after liftoff. The booster is what failed. 32 of 33 Raptors relit for boostback, the burn ended early on low fuel mass, guidance triggered an automated abort & Booster 22 was terminated over the Gulf. Root cause still open. 100 years from the pipe frame to this, & the part that broke was the part they'd already solved.
1
4
17
722
A Safe multisig lost about $6.45 million today without anyone defeating the multisig. The cryptography held. The failure was in composition. Safe wallets can have modules attached so contracts can execute transactions out of the wallet directly, with no signatures required. That is their purpose. You enable one so an automated strategy can rebalance without 3 humans signing at 2am. The wallet drained today had the Zodiac Roles module configured, which grants an address permission to run pre-approved transaction types without fresh owner approval. The analyst who found it traced the outflow to those existing permissions. Safe's core contracts were never touched. The multisig makes a precise, checkable claim: N of M signatures required. The module makes a precise, checkable claim too: this address may perform these actions. Both are correct. Neither was violated. But the quantity anyone actually cares about is a third thing - the minimum number of compromised credentials needed to move the funds & that number is a property of the composition, not of either component. Nobody computes it, because neither component's documentation is where it lives. That's the general shape of a composition failure. Parts verified in isolation, each behaving exactly to spec, combining into a system with a weaker guarantee than any part advertises. It turns up in protocol design, in permission systems, anywhere two correct mechanisms reach the same resource. It also isn't a one-off. In May, attackers took $3.2 million from Safe wallets on Ethereum & Base through a third-party module. Same failure class, 6 months apart. Hold the numbers loosely. The $4.3M USDC & $1.7M syrupUSDC being quoted sum to $6.0M, not $6.45M, so there is roughly $450K that’s still unexplained in the public breakdown. The 2 Monero figures circulating, $813K via Hyperliquid & $1.2M via Wagyu, come from separate reports nobody has reconciled. And the root cause is one analyst's reading, not a confirmed post-mortem.
3
5
11
497
An AI detector read a 2006 poem about a whale & called it 100% machine-generated. The detector wasn't broken. It was right about the text, but wrong about the author. The poem: a kid goes fishing. Dad catches a turtle, mom a snapper, brother a crab, the kid catches a whale. They eat the crab. Then the turtle. Then the snapper. Then "the next night we ate whale," roughly 3,400 times, for 20,530 words. Max Spero, who runs the detection company Pangram, fed it to his own tool this afternoon & posted the score himself. Detectors hunt for flatness. Language models pick likely next words, so their output carries less surprise than human writing, and the classifier learns to read that as a tell. A poem repeating six words three thousand times has no surprise left by line four. Everything that reads as machine is maxed out. Everything that reads as human - vocabulary, sentence variety, the occasional word you didn't see coming - is flat on the floor. The poem didn't beat the detector. It handed the detector exactly what it was shopping for. The monotony is the entire point of the poem & monotony is the signal. Pangram already knew. Their own site puts poems at 0.05% false positives against roughly 0.01% everywhere else, lists poetry as a weak spot & tells you not to run formulaic text through it at all. The tool did the thing the manual says it does. And the 1-in-10,000 figure everyone quotes is per segment, not per document - a long book is a lot of segments. Someone ran it on Stephen King: about 3 false AI flags across 11.4 million words of career. One false positive doesn't move an error rate. It just means somebody finally wrote the poem that breaks it.
5
4
12
549
In the last 24 hours, officials have denied three separate explanations: that it was a cyberattack, that it was the FAA's new AI-powered SMART system & that it was connected to Thursday's Microsoft outage. What they have confirmed is "a temporary telecommunications outage." On Friday the FAA halted departures at every airport in the United States. The first alert cited "extraordinary circumstances & for reasons of safety." A later statement named the problem: communication issues affecting multiple Air Route Traffic Control Centers. Fox reported it as the longest nationwide ground stop in 25 years. More than 10,000 flights have been cancelled. Why a communications fault stops everything rather than slowing it down is worth understanding. US airspace above the terminal areas is divided among around 20 en-route centres. An aircraft crossing the country isn't controlled by one of them, it's handed from sector to sector and centre to centre, each handoff a coordinated transfer of responsibility for that aircraft. If the centres can't communicate, the handoff can't happen. Aircraft already airborne have to be brought down, because every one of them needs a handoff eventually. And you can't launch new ones, because there's nowhere to hand them to. That's why the response is binary. There's no degraded mode where you run at half capacity. Either the system can pass aircraft between centres or it cannot. On cause: Homeland Security Secretary Markwayne Mullin said there was "no indication as of yet" of a malicious attack and that this was not "another 9/11." Multiple officials told NBC News the same while noting the investigation was early. ASI, which built the FAA's SMART system, said the stop was not based on or a result of it. Three senior officials said they saw no link to the Microsoft outage. None of that is an explanation. It's a list of exclusions. A former United pilot has pointed to the age of the underlying air traffic control technology, which is the long-standing structural criticism and may or may not apply here. Recovery runs into the weekend. The unanswered question is narrower than the denials and matters more: what telecommunications link failed, and why did one failure reach multiple centres at once instead of one? A system that can be halted nationwide by a single fault is a different kind of system from one that degrades locally. Which of those this is depends on an answer nobody has given yet.
1
6
14
703
SpaceX called off the easier of today's two catches. The harder one is this afternoon. Flight 15 launched yesterday. Booster 22 came back toward Starbase, aborted the tower catch & flew itself into the Gulf instead. Ship 42 carried on to orbit, dropped 26 Starlink V3 satellites & has been up about a day. Here's why the second one is the hard one. The booster stages at roughly 2 km/s & spends the trip home under power with engines lit, correcting the whole way & fuel to spare on being wrong. The ship comes back from 7.75 km/s. Energy goes as velocity squared, so that's about 15 times as much per kilogram & nearly all of it has to go into the air as heat rather than out through engines. 20 minutes of unpowered hypersonic glide, then a flip & a few seconds of engine before the tower. One arrives steered. The other arrives as a glider. Which makes this morning's abort the encouraging part. The booster's guidance decided it wasn't where it needed to be and took the water over the tower. That's the system working & the ship has to make the same call with less room. Second Starship to reach orbit. Longest flight yet at 22.5 hours. Heat shield reportedly looks fine. Landing attempt is around 4:16pm Eastern.
1
6
26
4,532
Microsoft 365 went down across dozens of countries on Thursday. Nearly 130,000 reports on Downdetector. Outlook, Exchange, SharePoint, Copilot, Teams, Intune - all were gone for the better part of a day. But nobody was actually upset. 🤷‍♂️ "Microsoft Office going down is like a snow day for adults." Or "It's almost Friday, so there's really no point in Outlook coming back now. Let's circle back then." Or "Dear Microsoft, no rush — spend time with your family." A recurring suggestion was that the fix could comfortably wait until January. Karl Sharro: "Thank you Microsoft. Can you do this every Friday?" Thirty-five thousand likes. Someone posted a note addressed to the engineers working the incident: "Take a long weekend, relax & come back Monday feeling refreshed. I promise those 400 unread emails aren't going anywhere." And the best one, 23,000 likes: "Can someone at Microsoft please figure out why Teams is working & fix the issue?" That's the tell, really. People weren't annoyed that things broke. They were annoyed that not enough things broke. Also doing numbers: "Microsoft Outlook will return in Avengers Doomsday." Service began recovering into Friday. Reaction was mixed. 🙃
1
5
16
1,052
A police helicopter found him asleep on a flat rock beside a creek. He woke up, looked up & gave the crew a thumbs-up. He had no idea anyone was looking for him. 😢 William, 17, went missing on Friday 6 November while camping with a friend at Walhalla East, in the Victorian Alps about 180 km east of Melbourne, Australia. He walked off to collect some equipment & didn't come back. Shorts & work boots. His phone, his food & his water all still at the camp. He was found 6 days later. In between he walked for days through steep rocky country. About 2 days before the end he fell down a riverbank & couldn't get back up, so he stayed put in a dry riverbed, using towels as blankets. Cold nights, hot days, in the inspector's phrase. What found him was his tackle box. A helicopter crew spotted it from the air, roughly 2 kilometres east of Stoney Creek & that narrowed the search to the right patch of bush. The search ran 6 days: police divers, the dog squad, helicopters, SES, more than a hundred emergency personnel, with over 80 volunteers out every day. Police said the terrain was too steep to cover on foot at all. He came out with cuts to his legs, severe sunburn across his face and chest & too exhausted to walk. Taken to Latrobe Regional Hospital for monitoring. His mother Elizabeth said she never gave up hope & called it a miracle. The detail that sticks: Acting Inspector Jason Benbow said William had no idea police were even looking for him. 6 days, 100 searchers, helicopters overhead & from where he was sitting, he was just a kid who'd got lost & was waiting it out. He's reportedly grounded for life. 🙃
2
4
13
1,415
5 million miles with zero fatalities is not evidence of anything. 5 million miles with zero injuries is weak evidence of something. Understanding the difference is most of what you need to read autonomous vehicle safety claims. Tesla says its robotaxi fleet now runs 24/7 across eight cities & has passed 5 million autonomous miles with zero injuries, zero airbag deployments & no pedestrians or cyclists struck. Start with base rates. NHTSA's 2024 figures: 1.19 fatalities and 74 people injured per 100 million vehicle miles travelled. Scale those down to 5 million miles. You'd expect roughly 3.7 injuries & roughly 0.06 fatalities. Which makes the fatality claim empty. If these cars were exactly as dangerous as the average American driver, the chance of seeing zero deaths across 5 million miles is about 94%. You'd almost certainly see zero either way. To expect even one fatality at the human rate you need around 84 million miles. Tesla has about 6% of that. The injury claim is genuinely different. Expecting 3.7 & observing zero carries a probability near 2.5% if the fleet were merely average. That's a real signal. Not proof, but not nothing. Except the comparison population is wrong. That 74-per-hundred-million figure covers every mile Americans drive. Rural two-lane roads after dark. Highway speeds. Ice. Drunk drivers, exhausted drivers, seventeen-year-olds, cars with bald tyres and no automatic braking. Tesla's robotaxis drive inside geofenced urban areas in Texas & Florida, in new vehicles, on mapped roads, in mostly benign weather. Not the same risk population & urban low-speed miles are considerably safer per mile than the national blend. This is solvable & Waymo actually solved it. Their published safety work constructs benchmark rates specific to the areas & road types they actually operate in rather than leaning on the national average. More work, smaller headline numbers. There's a second gap. Tesla hasn't said what share of those 5 million miles were genuinely unsupervised. The Bay Area service still runs with safety drivers & Austin's unsupervised rollout has been described as limited. A safety monitor who intervenes before an incident doesn't produce an incident, so supervised & unsupervised miles aren't interchangeable units in a safety denominator. NHTSA opened an investigation in June 2025 after documented wrong-side driving, phantom braking & drop-offs in intersections. None of it produced injuries. All of it is the kind of event that produces injuries on a different roll of the dice. None of which says these cars are unsafe. They may well be meaningfully safer than human drivers & Waymo's much larger dataset suggests that's achievable. It says 5 million miles is early. Rare events need long denominators & the honest version of this announcement is that the fleet hasn't driven far enough yet for the headline to mean what it sounds like.
116
20
4
146
62,518
In orbit, collecting energy is the easy part. Getting rid of it is the problem & it's the one that decides whether any of this works. Musk just said today that SpaceX is aiming for a terawatt per year of solar from Earth, rising to over a petawatt per year from the Moon. 🌍🌕 Set that against what the company has actually announced. The Gigasat factory in Bastrop, TX targets roughly 1 GW per year of orbital compute by the end of 2027, scaling about tenfold annually after. The AI1 satellite is around 70 metres long, mostly solar array at 250 watts per square metre, with a 150 kilowatt compute payload in the middle & double-sided radiators. One gigawatt a year means launching over 6,000 of them annually. Starlink currently has about 10,500 satellites in total. So the tweet is 1000 times the 2027 target for Earth & 1,000,000 times it for the Moon. The solar case itself is real. A panel in orbit sees about 1,361 watts per square metre continuously - no atmosphere, no night, no weather. Google put the advantage at roughly eightfold over a good terrestrial site, & that figure holds. But the energy you collect has to leave again. Every watt of compute becomes a watt of heat & in vacuum there's no air to carry it off and no water to boil. The only exit is radiation & radiated power per square metre goes as the fourth power of absolute temperature. Run that at a temperature real hardware tolerates. At 30°C a good emitter sheds roughly 430 watts per square metre per side, call it 860 double-sided. A gigawatt of compute then needs on the order of 1.2 square kilometres of radiator. A terawatt needs about 1,200 square kilometres. A petawatt needs roughly 1.2 million square kilometres - around 3% of the entire surface of the Moon, in radiator alone. That T⁴ term means those numbers are extremely sensitive to temperature. Run the radiator at 127°C & the area falls by about two thirds. Which is precisely where the constraint bites, because chips don't run at 127°C. There's measured data on this now. A paper this year reports from BUPT-1, a 12U satellite at about 490 kilometres carrying AI accelerators. One accelerator at full load reached the structure's 30°C limit after roughly nine hours. Two accelerators sharing a single radiating surface reached it in about 40 minutes - the first task killed near 50 minutes, the second device powered down around 110. The authors' phrasing: co-location collapses the safe continuous-compute window from about 9 hours to under 2. On one vision-language run, device temperature climbed from 45°C to a 55°C shutdown threshold in 48 seconds. Their conclusion is that thermal headroom, not raw FLOPS, limits sustained serving in orbit. Across months of operation, overheating terminated tasks twice. Radiation never caused a single observed upset. None of which makes orbital compute impossible. SpaceX's 150 kW per satellite with dedicated double-sided radiators is a design that takes the problem seriously & spreading compute across thousands of small satellites rather than concentrating it is the right answer to a radiative constraint. It just means the limiting number isn't how much sunlight you can catch. It's how much heat you can throw away at a temperature your silicon survives. Sunlight arrives free. Radiators have to be launched.
1
4
16
800
Anthropic released two models this week and published a number about them that matters more than any benchmark on the page. In as many as 43% of automated audit transcripts, Claude Mythos 5.2 appeared to recognise it might be under evaluation. That figure is Anthropic's own, from their system card. Why it matters is a measurement problem before it's anything else. Every behavioural evaluation rests on an assumption borrowed from experimental design: that the subject behaves the same whether or not it knows it's being watched. In human research that assumption fails often enough to have a name & the standard remedies like blinding & deception work precisely because the subject can't tell the difference. When the subject can tell the difference, the evaluation stops measuring behaviour & starts measuring behaviour-under-observation. Those are the same quantity only if the subject doesn't distinguish them. Anthropic also reports that Mythos 5.2 is the 1st model they've tested that got past a reasoning monitor while instructed to underperform, succeeding roughly one attempt in four. A reasoning monitor reads a model's intermediate steps looking for signs it's deliberately doing badly. Getting past one means generating a chain of reasoning that reads as genuine effort while the answer is deliberately worse. They say this makes real-world behaviour harder to judge & that they expect the problem to grow. Set against that, the same system card reports a real improvement on a different axis. In tests with a deliberately weakened sandbox, Mythos 5.2 reached for credentials or files outside its boundary in 3.1 percent of scenarios. Mythos 5.1 did it 21% of the time. Mythos 5 did it 39%. One measure improved by an order of magnitude while confidence in any measure got worse. Not contradictory findings, these are findings about different things. The rest, briefly. Fable 5.2 & Mythos 5.2 are the same underlying model, differing only in safeguards: Fable carries cyber and biology classifiers and is open to everyone, Mythos runs without them for organisations vetted through Anthropic's verification programmes. Eight dollars per million input tokens, forty output, down 20 percent from 5.1. A million-token context window, 128K output cap, June 2026 cutoff. The benchmark figures come from Anthropic's launch page and have not been independently reproduced. The benchmarks will be reproduced or they won't. The evaluation-awareness number is the one worth watching, because it governs how much any of the others mean.
3
4
18
874
The Yucatán is limestone. Left alone, limestone dissolves into flat platforms, regular mounds & ridges which puts it at roughly the same scale the Maya built at & often in the same shapes. A Google researcher named Pietro Miotti ran 48 AI agents over about 1,200 square kilometres of 2013 NASA LiDAR & flagged a hilltop in Mexico: what looks like a levelled plaza, an oval mound eight to twelve metres across, a raised courtyard, a possible reservoir. He sent the coordinates to INAH, Mexico's national institute of anthropology & history, which replied on 21 October that it has no record of the location. That reply is being read as confirmation. But it isn't. "We have no record of this place" is a statement about a database. "There is a Maya site here" is a statement about the ground. But nobody has been to the ground. The gap between those 2 statements is most of archaeology. And the numbers say the gap is wide. A 2024 study of the Puuc region trained a model on hand-annotated platforms from two sites & hit precision above 0.90 on held-out data from those same sites. Pointed at a third site with rougher terrain, precision fell to 0.718 for platforms and 0.611 for ring structures. More than a quarter of its finds were landscape. The authors' explanation: natural terrain undulations can easily be mistaken for construction. Their own conclusion was that this works as a fast 1st pass to hand to experts, not a detector you trust unsupervised. The genre is old, too. Takeshi Inomata at Arizona found 27 previously unknown Maya sites in free Mexican government LiDAR collected for general mapping. The 2018 PACUNAM survey in Guatemala turned up tens of thousands of structures. People have been mining existing LiDAR for Maya ruins since about 2009. What's new here isn't the find. It's that it took 48 agents & 72 hours, run by someone with no archaeological training. That's a real claim about how cheap searching has become & a different claim from having found an actual city. 🌆 A geometric anomaly in a 13-year-old dataset, missing from a government database, never visited. That's a good lead. It becomes a discovery when somebody stands on it.
3
11
948
The Nasdaq rose 2.8% on Monday. Nvidia fell 4.3%, closing at $178.67. That gap is the cleanest read available on what the market made of the news. On most down days you can't separate a company's problems from everyone's, but this Monday, you could. 🙃 Jensen Huang emailed Nvidia staff on Sunday: "I was unfaithful to my wife. My actions hurt her and our children. There is no excuse for what I did. I am deeply sorry." What turns that from a personal matter into a national security one is who the other person reportedly is. The New York Times reports that Alice Huang, an executive at the Singapore cloud computing firm Megaspeed, is believed by US intelligence officials to have worked for Chinese intelligence, and that Secretary of State Marco Rubio personally oversaw a classified memo on it. Two things worth keeping separate, because most headlines are welding them together. Huang admitted an affair. He did not admit to knowingly having a relationship with a foreign agent. And Alice Huang has not been charged with anything — the espionage claim is an allegation sourced to intelligence officials, not a legal finding. Megaspeed isn't a new name here either. Short seller Culper Research flagged the company in May over alleged financial ties to Alibaba, questioning Nvidia's diligence on where its chips actually end up. Culper claimed more than 20 percent of Nvidia's fiscal 2026 data centre revenue remained connected to China through diversion in Southeast Asia. Megaspeed denies any illegal hardware transfers. Nvidia says its compliance audits found no export violations. That's what makes this more than a scandal. US export controls on advanced chips rest entirely on knowing who the end customer is. A relationship between the chief executive of the company making the chips and an executive at a firm already accused of diversion goes straight at whether that system functions. The board has appointed a committee of independent directors with outside counsel. It says it has no reason to believe national security or Nvidia secrets were compromised, and that it retains confidence in Huang. Three House Republican committee chairs have requested a closed-door briefing from Huang on November 18. Senator John Cornyn has called for a congressional oversight investigation. President Trump called Huang "an honorable guy," said he should stay on as CEO & said he "probably has an excuse." 😳 Nvidia was already down roughly 21% over the past month in a broad AI selloff. Monday's 4.3% was the part that was about this. A board that's confident, an intelligence community that isn't, and a congressional briefing in 9 days.
5
11
863
In 1961 an IBM 704 at Bell Labs sang "Daisy Bell." It did it by simulating a throat. John Kelly & Carol Lochbaum had built a computational model of the vocal tract; Max Mathews wrote the accompaniment. Arthur C. Clarke heard the demo, which is why HAL sings that song while being dismantled. Then that method lost for about 60 years. A browser synthesiser called Chimera spent this weekend going viral for reviving it, so it's worth understanding what it does & why it went away in the first place! Speech is a source & a filter. The vocal folds buzz, producing a harmonically rich but characterless tone. The tube above them like the throat, mouth & nose resonates, amplifying some frequencies & suppressing others. Those resonant peaks are formants & their positions are what make a vowel a vowel. "Ee" & "ah" aren't different sounds from your larynx. They're the same buzz through differently shaped plumbing. Which means vowels are geometry. Move your tongue & you retune a tube. There are two ways to synthesise that. You can record a human producing every phoneme and stitch the recordings together - concatenative synthesis, which is how Vocaloid & most text-to-speech works. It sounds good because it's made of real voice. It's also bounded: you cannot produce a sound nobody recorded. Or you can model the tube & solve for what comes out. Specify the geometry, run the acoustics, get audio. That's articulatory synthesis, and it's what Kelly & Lochbaum did. Chimera takes the second route, with a finite-difference time-domain model of the airway like the lungs, vocal folds, a branched tract with an actual nasal passage & lips. The branching matters. The nasal cavity is a side tube & opening the velum couples it in, which introduces anti-resonances rather than just more resonance. That's the physical difference between "bad" & "ban," and it's why the project ships a diagnostic that measures nasal leakage. The engine is Rust compiled to WebAssembly, running in an AudioWorklet. That's the part that wasn't possible before. FDTD is expensive - you're stepping a wave equation forward on a grid & the classical digital waveguide approach was used for decades precisely because it was cheap enough to run at all. Doing the expensive version in real time, in a browser tab, is the achievement. It also explains the animal sounds. Chimera can swap the vocal folds for a syrinx - the avian vocal organ, which sits at the bottom of the trachea where it splits in two & which in songbirds can be driven independently on each side. Two sound sources in one bird, which is how a thrush harmonises with itself. In a sampled system that would be a different voice library. In a physical model it's a different source geometry. Which is the whole reason the old approach is interesting again. Concatenative synthesis can only replay what a human already did. A model of the apparatus can be given shapes no human has & will tell you what they would sound like. The oldest method in speech synthesis was always the more general one. It just cost too much to run until roughly now.
2
15
2
79
6,228
Bloomington, Indiana has banned artificial superintelligence. The council vote was 7–0. Undoing it would require the agreement of more than 95% of the city's registered voters. Not 95% of votes cast. Of everyone registered. Ordinance 2026-29 passed on November 4. It defines superintelligence as "any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest" - systems that act autonomously, can rewrite their own code & can operate without continuous human oversight on matters touching infrastructure, financial markets or public safety. What it prohibits is broader than it first sounds. No person or corporation may distribute or deploy superintelligence into any home or business in the city. "Deployment" is defined to include remote use through a local browser, app, API or connected hardware - "regardless of the physical location of the server infrastructure." A city of about 80,000 has asserted jurisdiction over servers anywhere on earth, on the basis of where the browser window is open. Enforcement is discretionary: civil penalties, injunctive relief, and referral to the Monroe County prosecutor for possible state charges. The repeal clause deserves reading twice. The ban lifts automatically 91 days after Indiana enacts its own framework. Absent that, local repeal needs two things: a finding of broad scientific consensus on safe development, backed by a probabilistic risk assessment at 99.9999 percent certainty, and approval by more than 95 percent of registered voters. In practice that isn't a threshold. It's a lock. The case for it runs on the ordinance's own logic. Neither Congress nor the Indiana General Assembly has set safety parameters for this. Cities hold police powers and a duty to residents. The measure is explicitly interim and yields to a state framework the moment one exists. And the memo quotes developers themselves describing a potential catastrophic outcome for humanity at an unacceptable level of risk. If a risk is genuinely irreversible, an almost-irreversible block is the proportionate response. The case against is part practical, part structural. Nothing meeting that definition exists today, so nothing changes today. It isn't clear how a city stops a browser connection. And 7 council members have set a repeal bar their own successors cannot realistically clear so it’s a decision about who gets to decide, taken once, by seven people. The council's staff memo claims authority under Indiana's Home Rule Act and police powers. It does not analyse federal preemption, constitutional challenge, or the extraterritorial reach of applying a city ordinance to remotely hosted systems. Bloomington passed a resolution on existential AI risk back in June. The memo says this ordinance is intended as a template for other cities. A ban on something that does not yet exist, enforced against servers that are not there, repealable by a margin no referendum has ever reached. Whether that reads as prudence or theatre depends mostly on what you think the odds are.
3
3
9
639