8 posts from the last 24 hours on temperature2.

Tao says AI proof-chasing is now actively harmful

Terence Tao posted October 10, 2026 slides from his Caltech lecture arguing that optimizing AI for open-problem solves is now actively harmful to mathematics.

McKinsey sees 11 million US job switches by 2035

McKinsey partners wrote on Oct. 11, 2026 that 11 million Americans may have to change occupations by 2035, 3.6 times the usual pace, even as total jobs rise.

This week in tokens: accept the harm, rehearse the day after

OpenAI and Anthropic put a catastrophe on a six-to-twelve-month clock, Altman said to accept some harm, and the shipping and the fundraising did not slow down.

China will adapt jobs to AI after 10.52 million hires

Li Zhong said on Oct. 10, 2026 that China will shape jobs around AI, as urban hiring hit 10.52 million in nine months, 87.7 percent of the annual target.

Third strike shuts Yandex's Vladimir data center

Yandex said on October 11, 2026 that a drone strike shut its Vladimir data center, the third of five large sites hit since Thursday, including Sasovo, home to two of its three AI supercomputers.

Signals: lab arms, 146 ms clicks, small auditors

The Register on Booz Allen's eight industrial scenarios, a Hugging Face post's 146 ms click, and CNBC on METR's $71 million and fewer than 50 staff.

Microsoft-Decision-1 ranks sixth on the JevBench API board

Microsoft shipped Microsoft-Decision-1 on Qwen3.5-9B at $0.042 per million input tokens. JevBench ranks it 6th of 27 at a 0.46s median, not the 85ms on its chart.

Nadella: treat frontier models as insider risk

Satya Nadella told companies on October 10, 2026 to treat frontier models as insider risks and keep a human-held emergency brake, because a model provider's assurances do not transfer responsibility.

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

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