beat · 202 stories
Orbital hardware cannot be recycled, so SpaceX's planned AI satellite constellation would lift 1,000 tons of copper and 1% of the world's palladium and thallium into permanent loss each year.
Researchers rebuilt six patients' neck arteries on microfluidic chips, finding that local blood flow patterns, not just narrowing, drive clot behavior.
Amazon says onboard self-flying systems and FAA Part 135, the commercial-air-carrier license, are what turn a six-fold drone delivery expansion from a regulatory project into a schedule.
A 23,417-scan study finds depression isn't tied to a shrunken hippocampus, the brain's main memory centre, pointing instead to the frontal cortex (planning and impulse control), the anterior cingulate (decision-making and motivation), and the
Micro1's 80–90% margins come from reselling the same training data to many AI labs, a practice its 24-year-old founder says now stops at the Chinese border.
The top US reseller of Unitree's Chinese-made humanoid and quadruped robots says task-specific arms, not general-purpose humanoids, are the right fit for most buyers.
Ramp's Router routes prompts across OpenAI, Anthropic, DeepSeek, and others through one US-only API and is free through 2026, though its default policy logs inputs and outputs for a year.
The federal system answered Aaron Swartz's 70-GB JSTOR download with criminal charges; Meta's 81.7-TB book trove for AI training is a civil fair-use fight.
A Linux process now impersonates a MacBook Pro on Apple's device-identity service, pulling the same encrypted Find My key an iPhone would see.
A third-party tracker counted 54 cars and 170 rides without safety drivers in Austin. The market was already home to a much larger driverless fleet.
Astronomers modeled what SpaceX, Blue Origin, and Cowboy Space filings would do to twilight. The answer: a permanent fixture of reflected light.
A draft model is a small helper LLM that proposes tokens the main model then verifies, and Liquid AI's DSpark speculative-decoding configuration of that recipe trims function-calling latency by 57% on its 2.6-billion-parameter LFM2.5 base model.