NVIDIA pairs Vera Rubin with Groq 3 LPX, quadrupling token speed
Agent-focused inference infrastructure already adopted by SpaceXAI, CoreWeave, and Nebius unveiled at Hot Chips
Hyunkook Kim
Agent-focused inference infrastructure already adopted by SpaceXAI, CoreWeave, and Nebius unveiled at Hot Chips
OpenAI's push to bring agents to white-collar work beyond coding is exposing a stark gap between internal and external adoption
Amazon bought this robotaxi company in 2020. Drawing on primary sources, we trace the story from its 2014 founding through a self-certification detour, a federal exemption, five recalls, and its first paid rides starting August 10.
The deal bundles infrastructure, model development, and local deployment. First priorities are cybersecurity, voice, and an Arabic-language model.
It handles not just generated footage but real camera footage too, converting it to 16-bit EXR, ProRes, or HEVC. Available now for Max plan subscribers and above.
Google's flood forecasting model, which already sends alerts to 2 billion people across 150 countries, has now expanded into urban flash floods
Team and Enterprise admins can now control access to Claude's connected tools centrally through a single identity provider
The law-firm-focused model, built on Alibaba's Qwen instead of OpenAI, only pulls ahead when tested on its own proprietary data
The tool now uses Gemini TTS to create simulated users that automatically score multi-turn voice conversations
OpenAI released a video showing ChatGPT Work being used by voice, carrying a spoken idea straight through to a finished output
Galbot announced a world first at an opening-ceremony exhibition match. The English press release says 100 shots, the Chinese spec sheet says 20 — what exactly was counted still isn't clear.
666 teams and 2,056 robots at Beijing's second World Humanoid Robot Games — the winning 100m time dropped from 21.50 seconds to 8.64 in a year, but robots still can't slow themselves down without help