Blog
Writing about AI — how it works, where it helps, and what we learn building with it.
Latest posts
Europe's Under-13 Ban Got the Headline. The Kids Act Also Puts AI Companions Inside ItVon der Leyen announced the age limits in Strasbourg on Wednesday; the Commission presents the Kids Act on Thursday. The numbers are negotiable — the architecture underneath, which pushes age checks to the app store and pulls AI companions into child-safety law, is not.Artificial intelligence
Google Won the Voice Benchmark by 1.1 Points. It Undercut the Price by Seven TimesGemini 3.8 Live Extended Thinking takes the top spot on Artificial Analysis' speech quality index by 1.1 points over OpenAI's GPT-Live-1 Astra. The interesting number is the price: roughly a seventh of what the model it beat costs per hour.Artificial intelligence
Shanghai AI Lab Did Not Train Its Frontier Model. It Post-Trained Z.ai'sShanghai AI Laboratory released Atria Dawn Preview, a 744B agentic mixture-of-experts model, under an MIT licence — built on Z.ai's GLM-5.2 base rather than one of its own. The division of labour is the story, and export controls do not reach the layer it happened in.Artificial intelligence
The Serious Charge Against Moonshot Is Not Distillation. It Is the RoutingAnthropic's September threat report names seven Chinese labs and about 190 million extracted exchanges. Buried in it is a different allegation: that Moonshot and DeepSeek served Claude's answers to their own paying users. That one does not depend on unsettled copyright law — and the evidence for it is also its biggest problem.Artificial intelligence
Astra Does Less Work. That Is Both the Product and the ComplaintTwo weeks in, OpenAI's Astra is tied for the top of the independent intelligence rankings and gets there on roughly a third of the output tokens of its nearest rival. The same instinct to cut a task short is also what users are complaining about.Artificial intelligence
Amodei Asked the Industry to Coordinate. It Had Been Coordinating Since JulyThe Information and the Washington Post report that Anthropic, OpenAI and Google have met on a working-group basis since at least July about an industry standards body for frontier AI. Three labs, three incompatible designs, and an antitrust problem the proposal names out loud.Artificial intelligence
Z.ai Raised $5 Billion It Has Not Finished SpendingZ.AI told the Hong Kong exchange on Sunday it is raising about $5 billion — a discounted placement plus $3 billion of zero-coupon convertibles — two months after a $4 billion raise it has spent barely a third of. The structure says this is an option on compute, not a cash need.Artificial intelligence
Asia Put a Price on Pacing the FrontierAmodei called for slowing AI down, Altman agreed and shelved the IPO, and on Monday the Kospi closed 3.26 per cent lower with SoftBank off 11.2 per cent. The selloff was concentrated in memory and optics, which tells you what was actually being priced.Artificial intelligence
OpenAI Shipped Its First "Critical" Cyber Model. It Also Made It Harder to WatchAstra writes working zero-day exploits without human help — and moves part of its reasoning out of readable text. OpenAI says the safety numbers improved. Researchers say that is not the point.Artificial intelligence