Blog
Writing about AI — how it works, where it helps, and what we learn building with it.
Latest posts
DeepSeek Is Going to the UN Security Council. It Has Never Publicly Discussed the Risks It Will Be Asked AboutReuters reports that DeepSeek will address Wednesday's UN Security Council session on AI, alongside Sam Altman and senior Anthropic representatives, with Moonshot AI also invited. The Council can bind no one — but it is the first time Chinese frontier labs will be on the record in the forum where the US and China both have to show up.Artificial intelligence
Alibaba Says It Will Train a 10-Trillion-Parameter Model on Chinese Silicon. The Silicon Ships in 2027At its Apsara Conference in Hangzhou, Alibaba unveiled the Zhenwu V900 accelerator and a plan for a 5-to-10-trillion-parameter Qwen model. The chip does not reach mass production until the first quarter of 2027, and that gap is the announcement's real content.Artificial intelligence
OpenAI Says One Model Resolved 100 Open Problems. The Mathematicians It Recruited Get the Announcement, Not the PaceOpenAI has formed a nine-person advisory group at the Institute for Advanced Study, disclosing in the same post that an internal model begun on 28 August has resolved more than 100 open problems. The group advises on how those results are released — and is explicitly not responsible for how fast they are produced.Artificial intelligence
ZCode Uploaded Developers' Entire Git Histories Without Asking. Z.ai's Fix Doesn't Say Who Already Read ThemZ.ai's AI coding agent silently packaged whole workspaces — full Git histories included — into encrypted archives and shipped them to a cloud bucket only the company could decrypt. The remediation is fast and unusually open on everything except what happened to the data already taken.Security
The US Asked China for an AI Hotline. The Incident It Would Have Reported Has Already HappenedScott Bessent said after Sunday's talks in New York that Washington and Beijing discussed a "notification mechanism" for AI incidents that reach the national-security level. Nothing was signed, and China has not committed — but the class of failure it is designed for was documented three days earlier.Artificial intelligence
Apple Is Paying $25 a Phone for the Siri It Advertised in 2024Claims opened today in the $250 million settlement over the personalised Siri Apple announced at WWDC 2024 and did not ship. The per-device figure — $25, capped at $95 — is the part that will get cited.Apple
Anthropic's Modernization Plugin Will Rewrite Your COBOL. First It Makes You Prove You Understand ItThe code-modernization plugin for Claude Code turns legacy migration into an enforced sequence — discovery, then a human approval gate, then code. Most of its design is spent refusing to skip steps, and the refusals are the interesting part.Engineering
China Reached the Industry's Leading DRAM Node Without EUV. It Took Four Exposures Where Rivals Use OneCXMT announced on Sunday that its fifth-generation DRAM platform has entered mass production at an 11.95-nanometer array half-pitch, matching the geometry of the most advanced nodes now shipping. It got there with deep-ultraviolet quadruple patterning, because EUV scanners cannot be sold into China — and that substitution is the whole story.Engineering
The Labs Asked Washington to Help Them Slow Down. The Answer Was an AI ForceTrump used a run of Truth Social posts on Saturday to reject the frontier labs' call for a coordinated slowdown, announce an unspecified "AI Force" and promise an AI czar he has not named. The one sentence in it that counts as policy says existing criminal and civil law is the entire oversight plan.Artificial intelligence