AISep 22, 2026MIT Technology Review
Don’t be fooled by this summer of AI hype
Gebru and Bender argue this summer’s AGI theater — hacks, math “breakthroughs,” resignation fireworks — collapses when domain experts get time to look.
The file
Latest
Sourced AI briefs in publish order — newest first. Citeable originals live on the homepage under From Signal Desk.
AISep 21, 2026MarkTechPost
NVIDIA Introduces SoL-Pi: Auto-Research Loops That Cut Coding Agent Token Traffic by Up to 49%
NVIDIA’s SoL-Pi trims coding-agent token traffic nearly in half by evolving the harness — same agent brain, fewer wasteful round trips.
AISep 21, 2026MarkTechPost
SpaceXAI Releases Grok 4.7: A Larger Base Model at the Same $2/$6 Price as Grok 4.6
Grok 4.7 ships a larger base and longer RL run at Grok 4.6’s $2/$6 pricing — a price-performance shove, not a clean sweep of the leaderboard.
AISep 21, 2026Hugging Face
Transformers now runs llama.cpp quants
Hugging Face transformers can now load GGUF quants with ggml Metal kernels — local laptop models without leaving the Python API you already know.
AISep 21, 2026Hugging Face
Jun Kim, oMLX creator and maintainer, joins Hugging Face to support the MLX community
Hugging Face just hired oMLX’s creator — a bet that Apple Silicon local AI needs a funded maintainer, not another side project.
AISep 21, 2026OpenAI
Priorities and principles for effective third party assessments
OpenAI wants deeper independent scrutiny of safety cases and safeguards — on its terms, with scoped claims and room to remediate before publish.
AISep 21, 2026Ars Technica
Muse, Meta's extraordinarily privileged AI assistant, has a serious 0-day
A Muse zero-day lets any local app hijack Meta’s hyper-privileged agent — security theater meets ClickFix, and the agent becomes the malware.
AISep 21, 2026NVIDIA Developer
Simplifying Model Serving Across Multiple GPUs with NVIDIA TensorRT Multi-Device Integration in NVIDIA Dynamo-Triton
Dynamo-Triton can now own a multi-GPU TensorRT plan behind one gRPC call — latency drops hard, but you’re trading silicon for speed.
AISep 21, 2026MarkTechPost
AWS Strands Agents Team Releases Strands Harness: An Open-Source Agent Harness With 28% Lower Token Cost at Comparable Accuracy
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.
AISep 21, 2026NVIDIA Developer
How to Evaluate AI Agents From Tool Calls to Task Completion
NVIDIA’s eval guide says stop grading vibes and tool-call trivia — gate releases on whether the environment actually reached the goal state.
AISep 21, 2026TechCrunch
OpenAI forms math advisory group as its AI resolves more than 100 open problems
OpenAI set up a math advisory group at IAS after a Navier-Stokes splash — influence without a brake pedal on the research pace.
AISep 21, 2026TechCrunch
Meta’s Muse is outpacing ChatGPT’s early mobile launch
Apptopia’s early Muse numbers beat ChatGPT’s first 12 days on matched iOS turf — distribution muscle, not just model magic.