Fetching from the wire…
01
02
03
04
05
06
07
08
Models2026-08-12 · source-backed
The essay argues closed frontier labs risk concentrating too much power in too few hands, framing American open weights as the answer to Chinese open-source models (FT, corroborated by CNBC and Fortune). The sharpest lines: "I do not understand why anyone who believes that AI will eliminate most jobs and much of humanity's relevance would rush to build that future," and "The notion that AI is so dangerous that the only safe path is an extreme concentration of power seems inherently problematic." Meta abandoned the open-weights fight when it pivoted to a money-making model in December 2025. This is the re-entry. I don't think the essay's argument and Meta's commercial interest can be cleanly separated, and I'd say the same about Anthropic's position. Both things can be strategy and sincere.
Each link below shares sources, entities, or timing with this story.
This is a supply-chain fact, and most people are still treating it as a geopolitics argument. Sequoia published "America's Open-Model Paradox" on July 24 with the number that reframes the whole conversation: Qwen's share of open-model fine-tunes went from 1% in January 2024 to...
Mira Murati's lab finally shipped a full LLM, and it's Apache 2.0. Inkling is 975B total parameters with 41B active in a MoE configuration, multimodal on input (text, image, audio) and text out, trained on 45 trillion tokens. The context number is the fun part: 1M tokens in th...
His same-day hands-on leads with the clean Apache 2.0 license and reports positive results on code exploration and image description, flagging the GGUF build immediately with llama.cpp, MLX, and ExecuTorch integrations following. His read carries weight because he's spent a ye...
For about a year, "run your agent locally" meant accepting a model that couldn't reliably call a tool twice in a row. That excuse is gone. Meta Superintelligence Labs published Muse Glimmer today: a 29.6B dense causal transformer, 52 layers, 6,656 hidden dim, with a ~1.8B ViT-...
Huang met Senators Cruz and Warner on July 28 arguing for American leadership in open source and pointing to Nvidia's pledge of up to $500B in domestic AI infrastructure. He separately told YC Startup School the jobs-apocalypse framing is "exactly backwards", that AI kills tas...
Meta's flagship model "Avocado" delayed from March to May after testing showed it performs between Gemini 2.5 and 3.0 — behind OpenAI and Anthropic. Meta leadership reportedly discussed temporarily licensing Google's Gemini. Fortune ---
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.