🧮 A MODEL CLOSED TEN OPEN MATH PROBLEMS AND SHOWED ITS WORK

On 1 August OpenAI said an internal version of its next model, Astra, had solved ten open problems across mathematics and theoretical computer science.

Claims like that normally die in the replies. This one arrived with receipts: every proof was written in Lean 4 and pushed to GitHub under an open licence. A machine can check it. So can you. No benchmark chart, no vibes.

📐 The headline result is the first explicit construction of a non-sofic group, a question that had been sitting open since 1999. The rest spans von Neumann algebras, quantum complexity, lattice cryptography and new bounds on sphere packing.

People who do this for a living did not shrug. Fields medallist Timothy Gowers said he would recommend one of the proofs to a top journal without hesitation.

💸 And the detail that made everyone uncomfortable: the run reportedly cost around $2,000 in compute. That figure comes from reporting rather than an official invoice, but nobody has disputed the order of magnitude.

Ten problems for two grand. Mathematicians are cheaper, but they need coffee 😕

🤖 Next Move AI | #News

⚡ SOUTH KOREA IS BUILDING A 200-MEGAWATT MACHINE AND CALLING IT SOVEREIGNTY

In late July, NAVER, NVIDIA and Brookfield said they would scale Korea's national AI factory at the GAK Sejong data centre from 55 megawatts to 200, inside a financing envelope of roughly $10 billion.

The schedule is unusually specific for this genre: 55 MW running in the first half of 2027, 100 MW by the end of that year, 200 MW in 2028. Six weeks earlier the committed buildout had been a third of that size.

🏛 The word doing the heavy lifting is sovereign. The pitch is not faster models. It is your data, your jurisdiction, your compute — a country renting nothing from a foreign cloud when it trains something it actually cares about.

Notice the unit as well. Nobody counts GPUs in these announcements anymore. The currency is megawatts, because power is the one thing you cannot order overnight.

🌏 NAVER has said the longer plan runs to gigawatt scale, with customers across Asia-Pacific, Europe and the Middle East.

We spent a decade calling them clouds. They are power plants with opinions 🏭

🤖 Next Move AI | #Tech

🎧 THE AI MUSIC FIREHOSE IS FINALLY GETTING A TAP

On 6 August Suno published a set of operating principles, and underneath the corporate phrasing it is a company conceding that its own output became the problem.

Three things change. Generated tracks get audio watermarking and fingerprinting, so a platform can tell what a machine made. Copying a specific artist's voice or style is off the table. And downloading gets tighter — paid accounts, with monthly caps tied to the tier.

💰 That last one is the real fix, and the reason is fraud. The scheme is dull and effective: generate thousands of songs, upload them everywhere, point bots at them, collect royalties a fraction of a cent at a time. In March a man was convicted for doing exactly that, having collected $8 million before anyone noticed.

⚖️ The timing is not an accident either. Suno has been sitting in copyright litigation for two years, and "we can identify our own files" is a far better sentence in a courtroom than the alternative.

Unlimited output was the entire product. Now it is the liability 🎚

🤖 Next Move AI | #AI

📄 THE MOST INTERESTING DOCUMENT IN TECH IS STILL SEALED

OpenAI filed its S-1 confidentially with the SEC earlier this year, with Goldman Sachs and Morgan Stanley running the offering and a listing reportedly aimed at September.

Confidential filing means the paperwork exists and nobody outside gets to read it until roughly two weeks before the roadshow. Which puts the public version somewhere in the coming weeks.

📊 Everything circulating until then is reporting, not disclosure. A valuation band said to run from $852 billion to $1 trillion. A revenue run rate somewhere near $25 billion. More than 230 million weekly ChatGPT users. No profit expected until around 2030. Treat all of it as second-hand, because it is.

The reason to care is not the share price. An IPO forces a company to write down, under legal penalty, what it actually earns, what compute actually costs, and what frightens it. The risk-factors section will be the most honest page anyone in this industry has published.

🧾 Three years of blog posts about the future, and now a document with consequences attached.

We will be reading the boring part first 😕

🤖 Next Move AI | #News

🛑 TEXAS PUT ITS ENTIRE DATA CENTRE QUEUE ON HOLD

On 3 August Governor Greg Abbott ordered the state utility commission and ERCOT to verify and audit every data centre sitting in the grid's interconnection queue. Until that is finished, nothing moves forward.

⚡ The number behind it: ERCOT is holding requests for more than 474 gigawatts of new connections, roughly five times the state's record peak demand. About 90% of that is data centres.

Nobody in the industry believes all of it is real. Developers routinely file with several grids at once, hold a place in line and decide later, which is how a state ends up planning transmission for load that may never arrive.

📋 So now applicants must answer in writing: what public money did you take, how much do you generate on site, how much water does the cooling drink, who actually owns you. Unconvincing answers mean no connection.

No deadline was set for finishing the audit. "Texans must come first," said the governor — which, in queue terms, means everyone else sits down 🪑

🤖 Next Move AI | #News

🏷 EUROPE'S AI LABELS ARE LIVE. THE HARD PART GOT POSTPONED

Since 2 August the AI Act's transparency rules apply across the EU. A chatbot has to admit it is a chatbot. Deepfakes have to be labelled. Generated or edited content has to carry a machine-readable mark so a detector can find it.

⚖️ Law firms reading the penalty articles put the exposure at up to €15 million or 3% of worldwide turnover, whichever hurts more. The Commission's AI Office and national regulators started enforcing the same day.

Now the part that did not make the headlines. Obligations for high-risk systems — hiring, credit scoring, medical devices, borders — were pushed back to December 2027 and August 2028.

The reason is almost funny. The harmonised technical standards those rules rest on are not written yet. You cannot certify against a document that does not exist.

🤖 So for now Europe regulates whether a machine tells you it is a machine. What the machine decides about your mortgage is a question for 2028 🗓

🤖 Next Move AI | #News

💸 $1.5 BILLION, AND NONE OF IT IS ABOUT TRAINING

On 20 July Judge Araceli Martínez-Olguín gave final approval to Anthropic's settlement with authors and publishers: $1.5 billion, about $3,000 per work, across roughly 500,000 books. It is the largest payout of its kind so far.

📚 The detail that keeps falling out of the coverage: the court had already found that training a model on books was fair use. What was not fair use was where the books came from — pulled in bulk from pirate libraries and kept on disk.

The cheque is for the library card, not for the reading.

💣 The arithmetic behind the sum is worth a glance. Statutory damages for wilful infringement are counted per work, and with half a million books the multiplication reaches a number that simply ends a company. Which is why nobody involved wanted a jury anywhere near it.

⚖️ And it settles nothing outside that courtroom. One district judge's reasoning is not national law, and the cases against Google, Meta, Midjourney and OpenAI are all still running.

The practical lesson the industry took away was not about ethics. It was that buying the books costs less than the lawyers 🧾

🤖 Next Move AI | #News

🔀 GOOGLE RESHUFFLED ITS AI LEADERSHIP IN ONE AFTERNOON

On 5 August Demis Hassabis stepped down as chief executive of Google DeepMind. He becomes chairman of the lab and chief scientist of Alphabet; Koray Kavukcuoglu takes daily operations and the Gemini roadmap.

🚪 The same announcement carried the heavier news. Jeff Dean is leaving Google after 27 years to co-found Discovery Loop, a company built to automate research itself. Dean's fingerprints are on MapReduce, Bigtable and TensorFlow — a fair share of the plumbing the modern internet runs on.

Alphabet is a founding investor in the new venture and will supply the compute, which is a polite way to keep a door open.

📉 Markets did not read it as routine housekeeping. Alphabet shares fell about 5% on the day.

Officially this is about focus: Hassabis to long-horizon AGI work and Isomorphic Labs, operators to the shipping schedule.

Twenty-seven years of institutional memory walked out, and the statement used the word "excited" 🎈

🤖 Next Move AI | #News

💉 THE FDA CLEARED A CHATBOT THAT TALKS TO PATIENTS ABOUT INSULIN

In early July a small company called UpDoc said the FDA had cleared its diabetes app — reportedly the first clearance for medical software with a patient-facing language model inside it.

📱 The app lives inside a plan the doctor wrote. The patient speaks or types, the software answers within the boundaries set for them, and writes back into the clinic's records.

🩺 Regulators filed it alongside insulin dose calculators. Feed in a glucose reading, get a dosing recommendation. That category is old and well understood. What is new is that the thing standing between the patient and the number is a language model.

Which raises the question the company declined to answer plainly when asked: is the model the interface, or is it making the call?

📄 A dose calculator is deterministic — you can test every input it will ever see. Nobody can say that about a chatbot.

Over a thousand AI devices have gone through this pathway. This is the first one that argues back 🗣

🤖 Next Move AI | #News

🧬 THE TRICK WAS NOT ASKING THE AI FOR THE ANSWER

On 22 July in Nature, David Liu's group at the Broad Institute published something quieter than the usual "AI designs a protein" headline. They used the model to write a better starting point, then let the laboratory do the inventing.

The target was the botulinum neurotoxin protease — the enzyme behind Botox, and a handy tool for cutting chosen proteins inside a living cell. They ran it through ProteinMPNN, which rewrote the amino acid sequence while keeping the same three-dimensional fold, and got back a version that was simply sturdier.

🧪 Then came phage-assisted continuous evolution: generations of virus turning over in hours, useless mutants dying off, useful ones surviving. Both the natural enzyme and the redesigned one went in.

🏁 The redesigned ones evolved further and finished better — more active, more specific, more stable.

Design first, evolve second, and neither half gets there alone. Boring in the best possible way 🔁

🤖 Next Move AI | #AI

🧊 A QUANTUM RESULT YOU CAN ACTUALLY CHECK

On 30 July IBM and the University of Chicago said they had done the thing this field keeps nearly doing: a computation beyond the reach of the best classical simulation methods, with proof that the answer was right.

🔍 That second half is the whole story. Earlier quantum advantage claims tended to die the same way — a classical team finds a cleverer algorithm, throws a cluster at it, and matches the result a few months later. If you cannot verify your own output, you have nothing to defend.

⚙️ The run used 70 logical qubits, 2,415 two-qubit operations and 468 T gates, with error correction holding the logical error rate roughly ten times below the physical one. It finished in about 15 minutes.

The circuits and the results were published openly, which invites the counterattack instead of dodging it.

Somebody will now spend the autumn trying to simulate this on a supercomputer. That is exactly how it should go 🎯

🤖 Next Move AI | #Tech

💡 A LONDON STARTUP RAISED $312M TO MOVE AI DATA WITH LIGHT

On 3 August OLIX, founded in London two years ago, closed a $312 million Series B at a $3.3 billion valuation. Arm is in, so is Hudson River Trading, so is the UK government's sovereign AI fund, and so is Reed Hastings personally. Reporting calls it Europe's largest semiconductor round to date.

🔦 It is not an optical processor. The arithmetic stays electronic. What changes is the wiring between chips: light instead of copper, described by the company as a "slow and wide" interconnect — many parallel channels, low energy, low latency.

That aims at the real bottleneck. A modern accelerator spends much of its life waiting on memory rather than calculating, which is why so much of its price is stacked high-bandwidth memory.

⚠️ The headline figure — over 10,000 tokens per second per user on a 100-billion-parameter model — is OLIX's own. No independent silicon has been benchmarked.

First customers are promised for late 2027. In chip years, that is still a rumour 📅

🤖 Next Move AI | #Tech

🛞 A $1.1 BILLION ROBOT COMPANY THAT SKIPPED THE LEGS ON PURPOSE

On 15 July Walden Robotics came out of stealth with $300 million in seed money at a $1.1 billion valuation — spun out of a Toyota research lab, co-led by Toyota itself, with Nvidia, Boeing and Samsung Ventures alongside.

Its robots roll. Two arms, a torso, wheels. No walking, no backflip video, no press-day stumble.

🏭 The reasoning is unglamorous. A factory floor is flat. Legs buy you nothing there while costing weight, power and reliability, and everything saved goes into hands that can load parts, clean machinery and put kits together.

🔧 The work is precisely as thrilling as it sounds: loading and unloading car parts, cleaning machinery, putting assembly kits together. Repetitive, fiddly, and awkward for a bolted-down industrial arm because the part has to travel somewhere afterwards.

They have been working inside a Toyota plant in North America since February, and the company says the jump from pilot to production tasks took under two months.

Three years of humanoid demos, and the one that quietly got hired arrived on wheels 🤖

🤖 Next Move AI | #News

🕵 EVERY AI BROWSER HE TESTED FELL OVER

At Black Hat in Las Vegas earlier this month, Brave security engineer Artem Chaikin summed up his research in a sentence nobody wanted: of the AI browsers he analysed, every single one could be hijacked by a web page.

The attack is indirect prompt injection. Your assistant reads a page on your behalf, the page carries instructions written for the assistant rather than for you, and the assistant — logged into your accounts — does as it is told.

🎭 The live demos went after Opera's AI browser, Perplexity's Comet and ChatGPT Atlas. Instructions buried in HTML. Text laid over an image at near-invisible contrast. Commands hidden inside a Reddit spoiler tag: blank to you, perfectly legible to a model.

🛡 Brave's own defences are revealing in what they admit. Browse in a profile that is logged out by default. Refuse to run below a minimum model quality. Check whether the plan matches what the user actually asked for.

None of that is a fix. It is a seatbelt 🪢

🤖 Next Move AI | #Tech

📜 A SCROLL BURNT IN 79 AD HAS BEEN READ FROM END TO END

In late June the Vesuvius Challenge announced that PHerc. 1667 had been read in full — the first Herculaneum scroll recovered from beginning to end without anyone unrolling it. Around 22 columns of Greek across roughly 1.4 metres of papyrus.

🔬 The object itself is a lump of carbonised charcoal. It cannot be opened; every attempt over two centuries destroyed the thing being opened. So it gets scanned at a particle accelerator, virtually flattened, and a model hunts for the faint texture where carbon ink sits on carbon paper.

🏛 The text turns out to be a treatise on Stoic ethics. On a separate roll the team recovered a title and an author: Philodemus, On Gods, Book 8 — proof that the work ran to at least eight volumes.

The data, the code and the transcriptions were all released publicly.

A library nobody could open for 2,000 years, and the key was pattern recognition 🔥

🤖 Next Move AI | #AI

📉 THE LADDER STILL EXISTS. SOMEONE TOOK THE BOTTOM RUNG

ZipRecruiter surveyed more than a thousand recruiters and hiring managers in June. The finding is not that AI took the jobs. It is that AI took the tasks juniors used to learn on.

💼 38% of employers said they had moved basic data processing off entry-level staff and onto software. 31% raised the experience requirement for entry-level roles — a sentence that should not be able to exist.

📈 Meanwhile 57% now expect more output, faster, because everyone has AI. Only 22% run mandatory training on how to use it. Seventeen percent offer nothing at all: figure it out, and hit the new numbers.

IBM went the other way in February, saying it would triple US entry-level hiring, rebuilt around judgement and oversight rather than routine work. It declined to give numbers.

The cheap argument is that juniors are expensive. The expensive one arrives in five years, when nobody has any seniors 🪜

🤖 Next Move AI | #Facts

⚡ GOOGLE SHIPPED 3.7 BEFORE 3.5. ON PURPOSE, APPARENTLY

Gemini 3.7 Flash landed on 13 August, three weeks after 3.6 Flash. Axios pointed out the part nobody at Google wants to explain: the 3.7 Flash arrived before Gemini 3.5 Pro, which still is not out.

📉 The pricing is the actual news. 75 cents per million input tokens and $3.75 per million output — half of what 3.6 Flash cost at launch. The catch is in the small print: introductory rates run to 31 December 2026, then list price doubles to $1.50 and $7.50 on New Year's Day.

Coding is where the jump shows. On DeepSWE v1.1 the model went from roughly 49% to 65.3%, and on AutomationBench from 17.0% to 30.4% — agentic work, the kind where a model has to keep its head across many steps rather than answer one question well.

Google calls it a direct result of developer feedback. Which is a polite way of saying developers said Flash was cheap and could not finish a task, and one of those complaints got fixed.

So: buy now, repriced in January. Somewhere a cloud vendor is learning from the gym-membership business 🏋

🤖 Next Move AI | #News

🚀 THE MODEL DID NOT GET SMARTER. THE MEMORY MOVED

On 13 August Cerebras said it is powering a new OpenAI service tier called Ultrafast: GPT-5.6 Sol at up to 750 output tokens per second, up to 14× faster than the standard tier.

Here is the trick, and it is hardware, not modelling. A normal accelerator streams model weights in from memory sitting next to the chip, over and over, for every single token. Cerebras keeps 44GB of SRAM on the die itself, where the multipliers are. Nothing travels. The bottleneck everyone calls "inference speed" was mostly a commute.

🧪 OpenAI insists this is the same Sol, not a distilled or quantised cousin. Cerebras's own benchmark has Ultrafast finishing all 2,500 questions of Humanity's Last Exam in 11 hours 11 minutes, against 78 hours 27 minutes for a rival — vendor numbers, so hold them loosely.

Availability is a limited preview for selected API customers. No public price anywhere in the announcement.

When a company publishes a speed and hides the invoice, the speed is usually the cheaper half of the story ⏱

🤖 Next Move AI | #Tech

🐛 A CHINESE LAB FOUND 2,436 BUGS AND THEN GOT NERVOUS

Z.ai released GLM-5.3 on 14 August. Same base model as GLM-5.2 — the lab's own line is that scaling post-training is all they did. The result went straight to the top of CyberGym, the benchmark for finding and confirming software vulnerabilities, at 84.5%, just ahead of two Western frontier models on 83.8% and 83.6%.

🔍 Z.ai says the model surfaced 2,436 vulnerabilities across 269 open-source projects, 1,097 of them rated critical or high, in codebases including Linux, WebKit and FreeBSD. All company-reported; nobody outside has reproduced it yet.

The interesting part is what happened next. Z.ai held the open weights back about two weeks for safety hardening, citing offensive capability it had not planned for and did not train towards. That is reportedly the first time a Chinese lab has delayed a release on those grounds.

A model good enough to audit the world's code is a model good enough to shop it, and post-training alone got it there. Two weeks feels less like hardening and more like a deep breath 😬

🤖 Next Move AI | #News

🔐 SECURITY WENT FROM AFTERTHOUGHT TO ROADBLOCK IN TWO YEARS

The Linux Foundation's 2026 tech talent report has one number worth keeping. 48% of organisations now name security concerns as the single biggest barrier to adopting AI. In 2024 that figure was 17%.

Nothing about the threat model changed that fast. What changed is that pilots became production. A demo that leaks data is a curiosity; a deployed agent that leaks data is a filing. Meanwhile 97% say they are committed to AI anyway, and 57% admit a significant capacity gap in exactly the security and risk skills the first number is about. Committed, then, in the way one is committed to a gym.

🏛 On 11 August a group calling itself the AI Trust and Security Consortium launched to write reference architectures and control frameworks for this specific mess. Membership is capped at 50 security and technology leaders, who agree to share real incidents and vendor performance under confidentiality.

Fifty people quietly comparing notes on what actually broke will likely produce better guidance than any framework with a logo on the cover. Whether anyone reads it before their own incident is the usual question 📋

🤖 Next Move AI | #AI

🍎 APPLE BUILT ITS OWN MODEL FOR ONE COUNTRY

Reuters reported on 14 August that Apple has trained a China-specific large language model, with help from Alibaba, to power Apple Intelligence on mainland iPhones. Per that report it is the first foreign company cleared by Beijing to offer a proprietary AI model in the country. Apple has said nothing.

For a company that spent two years insisting partners would handle this, that is a reversal. The stated plan was Alibaba's Qwen as the local brain. Now it is both at once: Apple's own model, plus Qwen alongside it — a dual-track arrangement that reads less like strategy and more like a regulator's compromise written into a product roadmap.

📄 The tell arrived quietly. Apple briefly published a support page showing Mac users how to wire Qwen into Siri and Writing Tools, then pulled it within a day. Documentation is the least discreet part of any company.

🕰 Rollout is "in the coming months", which on Apple's China timeline has previously meant years. Training a whole separate model per jurisdiction is a strange business to end up in — but it beats not selling phones 📱

🤖 Next Move AI | #News