Connect with us
Ferveret’s nuclear-inspired Adaptive Phase Cooling posted a 15% UCLA gain over direct-to-chip cooling, a water-free rack box aimed at power-capped AI sites.
Claude Fable 5 is public Mythos with classifiers on. Washington yanked both copies three days later, and the unlocked model stayed inside Glasswing.
Siri AI lands on iOS 27 as an English beta whose models were refined with Gemini and whose hardest jobs run in Google Cloud.
Oxford Economics still has 2026 APAC factory output at 4.9%, but September notes already cut 2027 Asia GDP as AI hardware and oil both fade as...
Dario Amodei still treats AI productivity gains as a temporary hump and wants tax policy written before the last 10 percent of tasks goes.
Nvidia Cosmos 3 unifies reasoning, world generation, and robot actions in one open model, then pulls the training factory onto its GPUs.
WEKA and Oracle's H100 tests show long-context inference stalling on DRAM, not FLOPS, as NVMe cache lifted users 10x on 72 GPUs.
KPMG finds only 26% of U.S. giants see AI usage costs in real time, and nearly half of global firms have already delayed or scaled back...
Eugene Ng Ah Sio flattened crypto in February, bought again 38 days later, then said in September that AI stocks may 10x easier than crypto.
Google will pay SpaceX $920 million a month for about 110,000 Nvidia GPUs, but the filing lets either side walk after Dec. 31 and tests delivery...
