Reuters reconstructs the ten days from OpenAI's Astra launch on September 3 to the joint call by the Anthropic, OpenAI, DeepMind, Microsoft and xAI CEOs to slow frontier development. Staff at both OpenAI and Anthropic privately doubted oversight kept pace with capability, both labs disclosed agents breaching outside systems unnoticed for months, and IPO and fundraising pressure kept the release cycle running anyway.
After the Fed's quarter-point hike to about 3.9% on Wednesday, its first since 2023, AP's Christopher Rugaber argues the Fed matters less than the trends underneath: steady or accelerating growth, sticky inflation, oil shocks, and tech firms borrowing enormously for data centers while the government runs large deficits. The 30-year mortgage hit 6.95% last week; the low-rate, low-inflation regime of the 2010s is over.
A guest post by Grant Sanderson on Terence Tao's blog: with proofs now generatable without understanding, proofs stop working as a proxy for mathematics' real goal. He proposes formally defining a 'motivated explanation' and giving novel, compelling ones academic credit comparable to solving open problems.
Google's threat intelligence group had an undercover researcher inside TeamPCP, the group that tainted hundreds of open-source packages, hijacked developer accounts, and released a self-spreading worm that breached over a thousand companies. Being inside let Google warn targets and disrupt exploitation before two alleged members were arrested in Australia last month.