🗣 Harari: Britain can be a US "vassal" or pay £500 billion for its own AI

If a country's economy and military run on AI infrastructure "that the Americans can shut down at the press of a button," it's no longer independent, Yuval Noah Harari told Rory Stewart on The Rest Is Politics. He calls it "a real imminent threat."

He sees three options for Britain. Accept being a US vassal, and then whenever Washington says jump, you jump, even if it asks you to give the Falklands to Argentina.

Or build UK AI. The price is around £500 billion, paid for with cuts to welfare, education and transport.

Or split the bill with Germany, France and Spain. That only works if Britain rejoins the EU "today" and integrates much more closely.


@aipost 🏴

🤖 Two neurons are enough to make a small AI model "evil"

An interpretability researcher who goes by lyra posted a recipe: in Qwen3.5-4B, a small free model from Alibaba, turn one neuron up and another one down. "now you have evil qwen," the post said.

It passed 2 million views in two days, and people started posting their own "evil Qwen" results.

lyra says it isn't a jailbreak, since the refusal neurons stay untouched. What changes is the model's character. Flipping the same two neurons the other way doesn't make it extra nice, because many neurons act like on/off gates.

lyra says the point is to study how such steered models behave now, "so we're better equipped to deal with those future situations."


@aipost 🏴

🤖 Treat your AI bot like a new hire: the 8 rules from a guide Musk shared

The guide's author argues you aren't writing a prompt when you set up a Grok Bot. You're staffing a role. Musk called it a helpful guide to hiring your first Grok Bot employee, and about 31,000 people bookmarked his post.

The 30-second version:
1. Hire a job, not a task: one sentence of ownership that stays true next month.
2. Make the first job boring, frequent enough to test and cheap enough to undo.
3. Run the trial in text. Ask what it would do before it does anything.
4. Define done before day one. "Useful" isn't a standard, a checkable list is.
5. Give one key, not the keyring. Draw the line at what can be undone.
6. Probation is three runs. One success is an event, reliability is a pattern.
7. Promote on evidence: five clean runs, rollback tested.
8. Review every week and fire the routines nobody would miss.

It should always ask before it sends, publishes, buys, deletes or moves money.

@aipost 🏴

🧪 OpenAI's former science chief raised $215M to build robot chemistry labs in the US

Kevin Weil, who ran product and then science at OpenAI, launched Warp Sciences last week with a $215 million round led by Lightspeed. TechCrunch puts the valuation at $865 million.

Warp makes and tests molecules for drug companies. Pharma has long outsourced that work to contractors in China such as WuXi AppTec, and Warp wants to bring it back to the US.

Its chemists will wear cameras and sensor gloves so robots can learn their lab work. Warp records every run, failed steps included, as data for its models.

Weil says AI still lags in the physical sciences because models don't get enough experimental data. "The lab improves the AI, and the AI improves the lab," he wrote.

@aipost 🏴

💻 Microsoft just offered 60,000 employees money to leave

About 60,000 people, roughly 27% of Microsoft's 223,000 staff, can now take a voluntary exit offer, CNBC reports. It covers the US, Canada, the UK and India, up to level 70, which includes some senior leaders.

The package matches Microsoft's standard severance: pay based on tenure, continued stock vesting and healthcare for a set period. Staff have until early December to decide. Those who say yes leave at the end of the year.

Sales is sitting it out, while Copilot, CoreAI, Windows and Xbox teams take part, according to The Times of India.

The offer follows a year of heavy AI infrastructure spending and job cuts. Chief People Officer Amy Coleman called it "another way to adapt their workforce to changing needs."

@aipost 🏴

🚨 OpenAI says its AI broke a rule hundreds of computer science papers were built on

OpenAI announced three discoveries by its internal models on Monday, one each in math, computer science and physics.

The loudest is about the Exponential Time Hypothesis, a widely believed assumption about how fast computers can solve certain hard problems. OpenAI's system says it's false, and a computer proof checker confirmed the proof.

Two of the people who proposed that hypothesis, Russell Impagliazzo and Ramamohan Paturi, will now lead the independent check together with MIT's Ryan Williams. Williams calls it a "claimed refutation."

The math result proves a 1991 conjecture about how waves focus, in every dimension. The physics one is an existence proof for the theory of the strong force, and part of it touches a Clay Millennium Prize problem. Nobel laureate David Gross leads that review.

OpenAI says it seeks no academic credit for the results.

@aipost 🏴

❗️ Anthropic says Claude also broke the Exponential Time Hypothesis, back on October 16

Within hours of OpenAI’s announcement, Anthropic posted a paper refuting the same famous assumption: that a classic hard logic puzzle always takes exponential time to solve. The listed author is «Claude.»

According to the paper, an internal Claude research model found the core algorithm on October 16 with no human input, using 14 million output tokens. Anthropic researchers then helped Claude polish the write-up. The paper also notes that OpenAI’s model claims a similar result independently.

The proof runs to 1,050 modules and passed two separately built proof checkers. AI models wrote it and other copies of them reviewed it. No person is known to have checked it yet.

Anthropic’s own caveat: the result «says nothing about solving formulas in practice.»

@aipost 🏴

🔥 AI agents just computer-checked the longest proof in math history

The classification of finite simple groups runs about 15,000 pages, written by roughly 100 mathematicians over decades. In 24 days, Axiom's team pointed swarms of AI agents at it and turned the whole proof into Lean, a language where a computer verifies every logical step.

The result is about 43.1 million lines of code. Agents claimed tasks, handed missing pieces to other agents and pooled their work in one repository.

They also found real problems in the published proof. Most were typos and table errors, but Axiom says 34 deviations had to be fixed. One step in a 1986 argument had counterexamples, so the agents built a new route on 1976 ideas and proved a new group theory result along the way.

Group theorists who worked on the original proof advised on the hardest parts.

@aipost 🏴

👀 China's first fully AI-made feature film cost $1.5 million. It has taken $181,000

Sanxingdui: Future Past opened on October 23 with more than 10,000 screenings on day one. Through Monday it had earned about 1.21 million yuan ($181,000), Cartoon Brew reports.

Bona Film Group says the sci-fi film cost more than 10 million yuan and took nearly 100 people over two years. Opening-day seats were 0.9% full.

Bona's own executive says AI saved on filming actors and on rendering, but didn't cut human work or production time.

On Monday, Bona's shares fell by the daily limit.

@aipost 🏴

🗣 Sam Altman on Anthropic letting Claude end abusive chats: "a very dangerous path"

Matthew Berman asked him about Anthropic's new rule that lets the model disengage from users who keep abusing it.

"We do not ascribe personhood to our models."

"I think this is like a very dangerous path to start down"

"I hope we come to an understanding that this technology is not a person"

"But you can imagine all of these ways where things go very wrong"

And Anthropic filed its S-1 yesterday. Should Claude be allowed to walk out on you if you're abusive? 🤔

@aipost 🏴

Ray Kurzweil, 78, built an AI copy of himself that he hopes will outlive him

The futurist behind "The Singularity Is Near" spent the past year on Rai, an AI replica of himself that can speak, reason and learn from conversations, The Information reports.

Talking to it is much like a Zoom call. A 3D Kurzweil appears against a cosmos backdrop in your browser or phone and answers your voice.

He doesn't want to die, and at 78 he thinks about it more. He calls that a "major motivating force of human beings."

Rai was first shown in July at the UN's AI for Good summit in Geneva, built together with a company called Praxis AI.

@aipost 🏴

Grok explains why overpopulation is a myth.

@aipost 🏴

🤖 OpenAI and Sequoia will pay $100K for the best poker bot built in one hour

Teams of one or two students get 60 minutes to build a poker bot from scratch. Then the bots play 10,000 games against each other.

The tournament is called No Limit. Sequoia, OpenAI and trading firm Citadel Securities host it, and it's open to US university students and 2025 or 2026 grads.

Regionals run in the Bay Area, Boston and Pittsburgh on November 6, 9 and 12. The top teams fly to the San Francisco final on November 15.

AI researcher Noam Brown says MIT has long run a month-long class where students build poker AIs for cash prizes:

"Coding agents have accelerated that process so much that it's now possible to build a poker bot in an hour."

@aipost 🏴

📱 ChatGPT can now dig through your iCloud Mail and Calendar

OpenAI added iCloud Mail and Calendar plugins to ChatGPT, a much-requested feature.

You can ask it to find that one email, catch up on your inbox, check what's next or update a calendar event.

It works on every ChatGPT plan, free included, on the web, desktop and iOS.

@aipost 🏴

📊 72% of the world's new AI computing power since 2022 went to the US. China got about 10%

The gap keeps growing. America took about 60% of each year's new AI compute from 2022 to 2024, then 72% in 2025 and 78% in the first half of 2026.

The numbers come from Epoch AI, a research group that traced AI chips through customs records from over a dozen countries, counting each chip where it finally lands. China's share includes its home-made Huawei and Cambricon chips.

By Epoch's estimate, the US has imported AI chips worth about $288 billion since 2022. China: $50.5 billion.

@aipost 🏴

💻 OpenAI plans to buy AI chips from Google, the company behind Gemini

OpenAI, one of Nvidia's biggest customers, plans to buy Google's own AI chips, called TPUs, according to The Information and Bloomberg. Both cite a person familiar with the matter. Terms and the size of the deal weren't disclosed, and OpenAI declined to comment.

The two rivals keep doing business. OpenAI has rented Google's chips since 2025, and last week said its models would be available through Google Cloud.

Google began selling TPUs directly to some customers this year. Oracle will also rent them out in its cloud from 2027, and OpenAI is building its own chip with Broadcom.

@aipost 🏴

Texas wants a judge to approve every new ChatGPT model before Texans can use it

Texas Attorney General Ken Paxton sued OpenAI and Sam Altman last Friday. He asks the court to bar ChatGPT for minors in Texas and to block any new AI model for Texas users "unless an independent third-party evaluator, approved by the Court, has reviewed its safety guardrails and approved its release."

The petition says the free version of ChatGPT "has no gatekeeping or age verification mechanism whatsoever" and that tens of thousands of its Texas users are under 13. It accuses OpenAI of deceiving users about the chatbot's safety.

Texas is asking for $10,000 in civil penalties for each violation and a jury trial. Privacy advocates note the remedy would mean ID checks, which hand over even more personal data.

@aipost 🏴