On the OnePlus 12's Snapdragon Hexagon NPU, llama.cpp achieved 4.5–12.5 tok/s depending on the model.
A multi-camera tracking system combines homography-based position prediction with sensor confidence.
Released a Claude skill that turns URLs into a citable LLM Wiki.
Pocket TTS has released a multilingual model supporting 6 languages along with an ONNX export tool.
Plano 0.4.22 adds a local TUI for checking cost and request counts, plus Ollama support.
Ran a Qwen3.6-27B UD-Q5_K_XL benchmark on an RTX 5090 with 19 runs and 93.9k tokens.
NVIDIA released an NVFP4 quantized checkpoint of the Gemma 4 26B model.
The Claude Code system prompt grew by 166 tokens in 2.1.124 and shrank by 87 tokens in 2.1.126.
To make use of AI-based design tools, building a sophisticated, machine-understandable design system is essential.
OpenWarp, which adds BYOP and multi-provider routing to a Warp-based terminal, has been released.
In a MacBook Pro M5 Max test, Gemma 4 31B outperformed in Pac-Man generation quality.
Meta's smart glasses footage review and the termination of its Sama contract have fueled a privacy controversy.
This analyzes why the Zig project bans LLM use and focuses on contributor growth.
OpenAI traced the rise of goblins and gremlins in ChatGPT to a Nerdy personality reward signal.
Introduced an AI multi-agent tool that generates, validates, and exports trading strategies using natural language.
IBM released Granite 4.1 3B, 8B, and 30B, with the 8B rivaling 32B-class models.
Applying Speculative Decoding to NeMo-RL boosted RL rollouts by 1.8x.
SKILL.md is not a prompt but a loader specification, and structure determines context cost.
Goodfire has released an early preview of Silico, an interpretability-based AI model design platform.
Cursor improved agent quality by customizing the execution system per model and using dynamic context.
The AI boom has triggered a super cycle in the memory chip industry, driving Samsung Electronics to record-breaking earnings.
Perplexity has added Teams, Excel, Workflows, and new data connectors to Computer for Enterprise.
SMG separated serving CPU work into a Rust gateway to eliminate the Python GIL bottleneck.
AWS Neuron has unveiled Neuron Agentic Development for developing NKI kernels for Trainium.