7 posts from the last 24 hours on temperature2.

OpenAI cuts Cursor off after SpaceX's $60B buyout

OpenAI says it'll stop feeding its models into Cursor on November 12, two weeks after Elon Musk's SpaceX closed its $60 billion buyout of the coding tool's parent company.

Unitree's G1 robot has a wormable Bluetooth root exploit

Security researcher Olivier Laflamme disclosed two root RCE chains in Unitree's G1 EDU humanoid, one reachable over Bluetooth with no pairing and able to spread itself between nearby robots.

Meta glasses now cut recording when you cover the LED

Meta is shipping a fix that stops Ray-Ban Meta glasses from recording once the privacy LED gets covered mid-clip, its second software patch for the loophole in two months.

Andreessen Horowitz bets $1.1B on AI's physical layer

a16z closes a $1.1 billion Machine Age Fund dedicated to chips, memory, power and data centers, betting the AI bottleneck has moved off the model layer.

OpenAI puts ads on ChatGPT's Free and Go tiers in India

OpenAI switched on ChatGPT ads for Free and Go users in India, its biggest market outside the US, with 50+ brands live and a self-serve manager coming September 4.

Nvidia pauses revenue-sharing deals with AI clouds

Nvidia halted parts of its $36B AI Compute Partnership after staff flagged antitrust risk in how it controlled cloud partners.

temperature2 ships /gpu/: what compute actually costs

A new section charting the two prices the industry runs on: GPU rental in dollars per GPU-hour for five accelerators, and blended token prices for four labs, with the depreciation curves, generational premiums and breakeven arithmetic drawn from them.

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