Spectra-FFT, a new optimizer presented on Hacker News, claims to cut AdamW optimizer-state VRAM usage by roughly 55% for full-parameter LLM fine-tuning, replacing the two 32-bit states AdamW keeps per…
Reflection introduced Beam, its first open-weight model, a sparse Mixture-of-Experts system with 501 billion total parameters and 23 billion active parameters built for coding, reasoning, and agentic …
StorageReview connected an ASUS ExpertCenter Pro ET900N G3 and an HP ZGX Fury AI Station — two NVIDIA GB300 DGX Stations — directly with two 400G direct attach copper (DAC) cables between matching QSF…
NVIDIA Inception startups iSono Health and Whiterabbit.ai are deploying FDA-cleared AI breast cancer tools to address screening and diagnostic gaps, with iSono Health's ATUSA wearable 3D ultrasound ca…
A developer known as Niko1221 has published Strata, a repository showing that the 125B-parameter Qwen 3.8 Flash Next model can run on a single consumer RTX 4090 at roughly 100 tokens per second. The s…
Typesafe.ai built MCJev, a Jev-style agent that makes each Minecraft PvP decision in a single forward pass using the DJev model, and its DJev-serve engine runs at about 24 ms per decision on NVIDIA GB…
A technical case study reports that a single NVIDIA RTX 4090 with 24 GB of VRAM can run a 125-billion-parameter Qwen 3.8 Flash Next model at roughly 100 trillion tokens per second, a throughput the wr…
NVIDIA released Physical AI Smart Spaces, a benchmark it calls the first to simultaneously provide large-scale, multi-class, multi-camera 3D perception data for indoor smart spaces, containing over 28…
A developer is building a production-grade 3D game engine for Windows, macOS, and Linux, using C++ and CMake with GLFW, NVRHI, Vulkan, glm, miniaudio, and Lua scripting. The engine is designed to be f…
OpenAI's transition to Dots marks a pivot to a subscription-based model for power users, separating its strategy from Meta's ad-driven Muse approach, according to the article. The piece argues that ev…
NVIDIA released DOCA Agent Skills, a set of pre-built capabilities that let AI agents interact directly with the BlueField Data Processing Unit (DPU) ecosystem instead of guessing at API calls. The sk…
NVIDIA confirmed at IFA 2026 that RTX Spark Windows PCs from Lenovo and Acer will ship in October 2026, pairing a 20-core Grace CPU with a Blackwell-class RTX GPU, up to 128GB of unified memory, and u…
SpaceXAI's Grok 4.7 tied MiMo-V2.6-Pro for first place on the new Artificial Analysis Cyber Index with a composite score of 56, edging past GPT-6 Luna at 53. The index, launched on or around September…
Moondream released Parakeet Redux, a speech-to-text model that compresses NVIDIA's 1.2 GB Parakeet 0.6B v3 to 178 MB via ternary quantization of the encoder weights, cutting Word Error Rate on seven E…
The open-source project Strata, from developer Niko1221, runs a 125-billion-parameter Qwen3.8 mixture-of-experts LLM on gaming-PC hardware by treating VRAM as a cache and keeping the full expert set i…
CoreWeave launched CoreWeave Forge at Fully Connected 2026 in San Francisco, a unified development layer that consolidates Weights & Biases, OpenPipe, and Marimo with new services including Model Dist…
Salvatore Sanfilippo, the creator of Redis, published DwarfStar (repo `antirez/ds4`), an MIT-licensed inference engine that runs DeepSeek V4 Flash, GLM 5.x, and DeepSeek V4 PRO entirely on consumer ha…
NVIDIA's IsaacTeleop framework converts XR hand tracking and motion-controller input into commands for simulated and real robots through a graph-based retargeting engine, according to a Marktechpost t…
President Trump signed Executive Order 14434 on September 29, 2026, requiring federal agencies to use "Super Intelligence" instead of "AI" and giving the new Super Intelligence Force 120 days to produ…
Anthropic shipped a built-in Claude Code plugin called "You should Know" that scans Claude's output for important information users might miss, enabled via `/plugin enable cc-plugin-you-should-know@bu…