The Frontier
The people best placed to judge whether AI can do mathematics keep landing on the same verdict — and one of them just proposed a clean test for when the machines would finally prove him wrong. Princeton number theorist Peter Sarnak has proposed an "AlphaZero Test" for AI mathematics
The Arena
The open-weight center of gravity has a new address: Alibaba's Qwen crossed 3 billion downloads in six months, and Hugging Face's summer ecosystem report crowns it the largest foundation in open AI. Alibaba's Qwen family has become the most-downloaded open model line
The Arena
Secondhand bookshops across the UK and Ireland say mystery buyers are clearing their shelves with erratic, theme-free bulk orders — and the trade increasingly suspects the books are being scanned and pulped to feed AI training pipelines. Booksellers in the UK and Ireland are reporting a wave of "strange&
The Frontier
Two stories in one morning, and both are about the machinery underneath the hype: a Genie-style world model small enough for a desktop GPU, and an industry quietly discovering that agentic AI is a CPU problem. A new open-source model lets you walk around a real-time, playable
The Everyday
Two stories about the human side of AI today: the field's most prominent advocate makes the case for what machines can't do, and a Meta patent hints at what they might soon do to bystanders' faces. Fei-Fei Li has a message for the doom-
The Guardrails
The anti-AI movement crossed a line this weekend — its first member went to prison — while a new study from Google researchers suggests safety training rewires chatbots in ways nobody planned. Wynd Kaufmyn, the 69-year-old retired teacher believed to be the first person jailed for protesting against artificial
The Arena
China's industrial AI policy machine keeps adding cities to its national pilot-zone program — this week it was Hefei's turn, and the designation comes with some of the most interesting numbers yet on domestic compute. China's Ministry of Industry and Information Technology has approved
The Stack
China's homegrown AI compute stack keeps closing the gap with the West — the latest sign: Huawei's Ascend platform ran Xiaohongshu's newest open model the day it shipped. Huawei announced that its Ascend Atlas 800 A3 servers and Atlas 900 A3 SuperPoD supernodes now fully
The Guardrails
Seoul's sovereign AI project — a government-funded contest to crown one "national representative" foundation model — heads into its second-round verdict under a cloud: foreign firms allegedly offered to juice the entrants' benchmark scores. Allegations that overseas companies offered to inflate benchmark scores for South
The Guardrails
OpenAI has quietly dismantled the unit whose entire job was deciding whether its own models could cause catastrophic harm — the clearest organizational sign yet of where safety sits as the lab races an IPO clock. OpenAI dissolved its "Preparedness" team at the end of July, the Financial Times
The Frontier
A research team from MPI for Intelligent Systems, ELLIS Tübingen, and ETH Zürich trained language models from scratch on nothing but U.S. elementary-school material — and found that no amount of post-training or clever prompting pushes them past the fifth-grade ceiling their pretraining data set. Researchers built
The Guardrails
A safety filter nobody noticed was missing, at the lab whose CEO calls bio-weapons risk the scariest thing in AI. Anthropic disclosed that its biological-weapons blocking classifiers were offline for nearly a year — from May 2025 through April 2026 — leaving roughly 133 million model exchanges from about 50,