Mlx.fast: qwen 3.8 27b triples speed on mac with multi-token prediction
Yukon Labs launches MLX.fast, a leaderboard that boosts Qwen 3.8 27B inference speed by +272.9% on Mac using Multi-Token Prediction.
Blog
The open weight illusion: why local AI depends on china
Qwen 3.8's release highlights the fragility of local AI. Between US closed-source hegemony and potential Chinese restrictions, access to consumer-grade models is at risk.
Bluesky: the decentralized challenge to x's dominance
Born from Twitter, Bluesky aims to reshape social media with the AT Protocol and user-selectable algorithms.
Closeall: the open source power tool for macos multitasking management
A lightweight, free utility to quit all running apps in one click while monitoring CPU, GPU, and RAM usage on macOS.
Beyond APIs: CLI-Anything and cli-printing-press Make Every Software Agent-Native
Two open-source projects are redefining AI-software interaction, turning any application into an agent-optimized tool via structured command-line interfaces.
Palmier Pro: The Open Source Video Editor Where AI Agents Control the Timeline
Palmier Pro is a macOS-native NLE that leverages the MCP protocol to let AI agents like Claude and Cursor trim, reorder, and generate video assets directly in the project timeline.
Beyond Steel: Fincantieri's Vision for the Intelligent Naval Ecosystem
Fincantieri is redefining shipbuilding by integrating AI, composite materials, and hydrogen propulsion into a connected digital ecosystem linking ships to shore.
I Built an Entire AI Newsroom. Here's How
A dozen AI agents and a single human supervisor to find, classify, select, group, write, illustrate, and publish dozens of news stories every day. In Italian and English.
DwarfStar (DS4): The Local Inference Revolution Is Just Beginning
Salvatore Sanfilippo, creator of Redis, has built a local inference engine for DeepSeek V4 that is shaking up the AI world. And with SSD streaming, it could bring near-frontier models to Macs with as little as 64 GB of RAM.
Follow our channels