Topics

A deep-learning system called WorldParticle simulates materials from sand to proteins in one architecture, potentially replacing many specialized simulators and aiding scientific discovery.

Google and SpaceX recently launched the first experimental satellite equipped with AI chips, testing whether orbital data centers could solve Earth's energy constraints. The project promises abundant solar power and reduced environmental impact, but faces significant technical hurdles including cooling, radiation, and data transmission.

Anthropic describes how notes, an offline review it calls dreaming, and wiki-like structures let Claude remember lessons from one session to the next.

Google will launch a prototype satellite on October 1 to test its Tensor Processing Units in space as part of Project Suncatcher. The mission will evaluate how AI chips perform under radiation, thermal extremes, and the physical stresses of spaceflight, marking the tech giant's first step toward potential orbital computing infrastructure.

Meta has revealed Prometheus, its most ambitious AI infrastructure project to date. Located in New Albany, Ohio, the supercluster is designed to train next-generation foundational models at an unprecedented scale. The first phase is already complete and was used to train the company's Muse family of models. At full scale, Prometheus will comprise roughly one million GPUs distributed across multiple interconnected buildings, all functioning as a single machine. The facility features advanced closed-loop liquid cooling and multi-generational NVIDIA hardware including GB200 and GB300 platforms.

Google's Project Suncatcher prototype satellite successfully reached orbit aboard a SpaceX mission to evaluate whether machine learning infrastructure can function reliably in space. The mission will test Tensor Processing Units under radiation, thermal extremes, and launch vibration conditions, with future milestones planned for 2027 including satellite laser interconnection tests.

EdgeCortix has introduced RAIDEN, a multi-die chiplet architecture designed to scale AI compute for robots, autonomous systems, and edge servers. The platform ranges from single-die to four-die configurations, with the flagship X4 offering up to 3.36 petaflops of FP4 compute and 256 GB of memory. The chiplet approach aims to solve the scaling limitations faced by physical AI systems that must perceive, reason, and act in real-time environments.