232 results for “IRST”
Moonshot AI has launched Kimi K3, a 2.8-trillion-parameter model with native multimodal capabilities and a 1-million-token context window. Open-model releases have mostly clustered well below the trillion-parameter class.
Lockheed Martin's Skunk Works division has completed a landmark test series in which an autonomous fighter aircraft executed 27 live intercepts relying. The strategic significance of the Have Heat demonstration extends well beyond.
Argonne and FRIB researchers have measured the krypton-88 neutron-capture rate for the first time, cutting its uncertainty from eightfold to threefold. Updated i-process models with the lower rate produce more strontium, closing a long-standing gap with old-star observations.
NVIDIA submitted its first MLPerf Inference v6.1 preview results for Vera Rubin NVL72, showing up to 3.7x higher throughput than GB300 NVL72 on Qwen3-VL. The numbers signal a leap for rack-scale inference on MoE models.
On 2 September 2026 the PyTorch Foundation released 2.14: nccl2 and NVGEMM land in-tree, fault tolerance becomes a first-class c10d feature, and Apple Silicon gains native linear algebra plus Metal kernels that close an 8.5× single-token decode gap versus the prior MPS path.
OpenAI released a major Agents SDK update adding native sandbox execution and a model-native harness for production agent workloads. The Python-first release includes configurable memory, Codex-like filesystem tools, and a Manifest abstraction for portable workspaces across seven sandbox providers.
On August 27, 2026, Google DeepMind announced the first double-blind evaluation of a frontier-class model, testing Gemini 2.5 Flash-Lite inside a cryptographic enclave. The approach lets outside auditors run confidential benchmarks without exposing prompts or model weights.
Janelia, Google Research, and collaborators published MaleCNS, the first complete male fruit fly central-nervous-system connectome (~166,000 neurons). Viral Minecraft + GPT-6 Astra demos are still early and unreleased — the open atlas is the durable story.
Nvidia has agreed to invest up to $3 billion in Lancium, the Texas-based power infrastructure developer behind the first operational site of the Stargate. Power has replaced silicon as the scarcest resource in frontier AI development.
First-day brief (Aug 5, 2026): four Google AI researchers leave to launch Discovery Loop. Superseded for depth by Brocker’s later PBC analysis — use this for the initial departure framing only.
A New Mexico judge has ordered Meta to pay $567 million into an abatement fund to address harms its platforms have caused to children, marking. The decision is the first in the United States to classify a social media company as a public.
Alibaba launched HappyShrimp 1.0 on August 17, an end-to-end AI music model that turns natural-language prompts into complete songs with vocals and a stereo master. The release gives creators a Chinese-language-first alternative to Suno and Udio with built-in commercial licensing.