beat · 163 stories
A Sept 29 executive order tells federal agencies to call AI 'superintelligence,' pinning the new label to the same 2020 statute that already covers spam filters and fraud detection.
Epoch AI's projection puts 1.9 billion concurrent AI agents on the table by 2027. The build that decides who captures the value is on the demand side, and it has not started.
OpenAI is reviewing 50 petabytes of records and has notified more than 100 organizations, including a New South Wales (NSW) portal that held historical bushfire data.
Propose-score-retrain is the new coding, and the human still sits at the four points where the loop cannot close itself.
The flight is solved, but the loop between flights, from charging to refilling to relaunching, is the hard part, and XAG (a Chinese ag-drone and robotics company), John Deere, and Hylio (a Texas-based spray-drone operator) are all racing to close it.
Airbnb CTO Ahmad Al-Dahle says 60% of the company's code is now AI-written and feature shipping is up 80%, but the real shift is a workflow collapse, not a productivity trick.
Four proposals, including a constellation of up to a million satellites, are now under FCC review; the hard part is the engineering and the multi-agency approval map, not the filing.
The coalition says the FAA's July environmental finding was too thin to clear the way for routine drone delivery under Part 108, the FAA's pending rule for routine commercial drone flights that operate beyond a ground pilot's direct view.
When a frontier AI lab fires safety staff and refuses to name the outside group or the data, it tests who sets the rules for independent evaluation.
A 367-person Phase 2 trial found Lilly's two-drug combo—an amylin (satiety-hormone) drug layered on top of tirzepatide, the GLP-1 drug sold as Mounjaro and Zepbound—beat top-dose tirzepatide alone on weight and blood sugar, with up to 27%
An arXiv preprint reports a 24.48% token cut on LongMemEval, a long-horizon AI agent memory benchmark, at the same accuracy level, by dialing a single budget knob that ranks how memory slots are filled.
A new agent harness splits "the model said so" into five testable steps. Its own pilot didn't show it was better.