Xiao Zha launches a high-profile battle with closed-source AI! Pushing 30B agent model, consumer-grade graphics card can run it, Yang Likun likes it

Compiled by Zhidongzhi | Edited by Eggplant | Cheng Qian Zhidongzhi reported on August 11 that last night, Meta launched the 30 billion parameter local Agent model Muse Glimmer and announced the opening of weights. This is also the first time that Meta has opened the model weights since the establishment of the Super Intelligence Laboratory. Subsequently, Meta founder and CEO Mark Zuckerberg published a long article explaining Meta’s super-intelligence route, and called for reducing U.S. restrictions on open source AI models. He criticized some AI manufacturers for controlling technology and users through closed models, and supported the continued implementation of chip export controls. He also predicted that Meta will soon open the weight of the latest flagship model Muse Spark 1.2. ▲Zuckerberg officially announced the opening of Muse Glimmer weighted concurrent long articles (Source: Meta positions it as the basic model of a local personal agent, which can be used in scenarios such as organizing files, managing schedules, drafting messages, local programming, and function calls. According to Meta, Muse Glimmer is optimized for continuously running local Agent workflows. The quantified size is less than 20GB. It can run on some Macs or PCs equipped with a single consumer-grade graphics card, and can perform tasks even without a network connection. Yann LeCun, Meta’s former chief AI scientist, founder of AI company AMI Labs, and Turing Award winner, applauded directly in Zuckerberg’s comment area and said (opening up the weight) is a great step. ▲ Comments by Yang Likun (Source: