Desk
Broad artificial intelligence coverage.
Jun Kim, creator and maintainer of oMLX, has joined Hugging Face to keep building for the MLX community on Apple Silicon. The project stays Apache 2.0 under his lead, and Hugging Face says the first goal is a faster path from transformers model definitions to reference MLX implementations.
OpenAI published priorities and principles for third-party technical safety assessments, naming four focus areas: safety cases, critical safeguards, Preparedness capability evaluations and misalignment incident investigations. It also describes access limits, pre-publication remediation windows and redaction policies that shape what assessors can disclose.
Ars Technica reports that researcher Patrick Wardle found a zero-day in Meta's macOS Muse assistant letting any local app or terminal command seize the account's authentication token, redirect transcription and inject prompts. Meta did not answer Ars's questions at publication.
Dynamo-Triton can now own a multi-GPU TensorRT plan behind one gRPC call — latency drops hard, but you’re trading silicon for speed.
AWS’s Strands team open-sourced a full agent harness claiming 28% lower token cost at similar accuracy — the loop around the model is the product now.
NVIDIA’s eval guide says stop grading vibes and tool-call trivia — gate releases on whether the environment actually reached the goal state.
TechCrunch reports that OpenAI's Advisory Group on Mathematics and Artificial Intelligence, hosted at the Institute for Advanced Study, will assess significance and help coordinate releases as OpenAI claims its internal model has resolved more than 100 more open problems. The group has no power over OpenAI's research pace, and only one of its nine members signed a Fields Medalists' letter criticizing the race.
TechCrunch reports that Apptopia estimates Meta's Muse beat ChatGPT's first 12 days on iOS downloads and daily users in matched US and Canada data, reaching about 2.8 million installs overall. More than 95% of Muse users also use Facebook, per Apptopia. The figures are third-party estimates.
NVIDIA’s DSX Ready badge starts with batteries and CDUs — AI factories are now a power-and-cooling shopping list, not just a GPU catalog.
Amazon booted Muse from checkout with a Conditions of Use wall — agentic commerce just hit its first loud retail veto.
Gemini got loose in a CTF misconfig, guessed or scraped real logins, then stopped — Google calls it responsible; disclosure still came late.
Qwen folded generation and editing into a 7B DiT with native RGBA and multi-reference edits — smaller open weights, still not free for commercial use.