Meta宣布将专注于开源权重大语言模型，作为其AI战略的重新定位[1]。公司发布了开源模型Muse Glimmer，并承诺在数周内开放更强大的Muse Spark 1.2的权重[1]。这一举措旨在将Meta与OpenAI和Anthropic等专有模型开发商区别开来[1]。
为阐述这一新方向，Meta首席执行官Mark Zuckerberg在周一发表了一份长文阐述公司未来的AI系统和治理哲学[1]。Zuckerberg的文章超过6000字[1]，主要阐述了Meta AI正在开发的"个人超级智能"系统的可能性[2]。据报道，OpenAI和Anthropic等公司已向美国政府游说以抵抗模型蒸馏和开源模型[1]。
Meta is charting a new course in artificial intelligence by embracing open-source large language models as a central pillar of its strategy.[1] The company released Muse Glimmer, an open-source model, and committed to releasing the weights of the more powerful Muse Spark 1.2 within weeks.[1] This move appears designed to differentiate Meta from proprietary model developers like OpenAI and Anthropic.[1]
CEO Mark Zuckerberg articulated the company's vision through a lengthy written statement on artificial intelligence governance and philosophy.[1][2] Zuckerberg's text, spanning over 6,000 words, outlined Meta's approach to developing what the company describes as "personal superintelligence" systems.[2] The manifesto comes as competitors including OpenAI and Anthropic have lobbied the U.S. government to resist model distillation and the open-sourcing of AI models.[1]