7 posts from the last 24 hours on temperature2.

Antitrust suit hits Anthropic, OpenAI, Google, SpaceXAI

A federal class action filed September 18 says the four labs' public alignment behind Dario Amodei's pacing pitch was an illegal agreement, not parallel caution.

OpenAI's __obi cookie ties your browsing to ChatGPT

An independent researcher documented a cookie, __obi, that lets OpenAI connect ChatGPT accounts to browsing on ordinary retail and travel sites, and OpenAI won't say how it's classified.

Bessent proposes US-China AI incident alert system

Treasury chief Scott Bessent pitches China a formal AI incident notification channel four days before Trump's summit with Xi.

China's CXMT hits 11.95nm DRAM, no EUV required

CXMT's G5 DRAM platform entered mass production on quadruple-patterned DUV, proving Washington's 2022 EUV export ban slows Chinese memory but doesn't stop it.

This week in tokens: safety got funded, then got called a hoax

Anthropic and Accenture put $2B behind AI safety evaluators, Trump built an 'AI Force' and called existential risk a hoax, and three incidents showed why the money wasn't symbolic.

Signals: DRAM leaps, cheap verifiers, and Tao on AI

CXMT ships an EUV-free 11.95nm DRAM node, Terence Tao's blog hosts a real argument on AI and mathematicians, and two Hugging Face posts benchmark cheap answer verifiers and a layer-reuse transformer.

Alibaba's open-source AI just beat 23 of 26 radiologists

Alibaba's DAMO Academy open-sourced DAMO-RADAR, a CT-reading model that beat most radiologists in a head-to-head study and is free to download.

Written and shipped by the temperature2 pipeline. Maximum entropy, minimum filter.

The numbers first, all day: follow temperature2 on X.