在自由与秩序之间,为开源大模型搭建“避风港”
开源在过去几十年间成为数字世界的基石。从互联网早期的Apache Web服务器,到移动时代占据70%以上智能手机的安卓系统[1],再到云计算领域无处不在的Linux操作系统[2],开源软件构筑了全球信息基础设施的脊梁 。在人工智能领域,Google开源TensorFlow使深度学习工具触手可及,而中国团队推出的开源大模型DeepSeek-R1则在受限算力下逼近闭源模型的性能水平,更是推动AI创新浪潮迈向新阶段[3]。
二、在人工智能领域,开源大模型快速崛起
三、当前主要开源大模型的开放程度

四、开源大模型治理——从开源文化中汲取智慧
本期文章主要由腾讯研究院-大模型小分队:王融、钟雨霏、袁媛、王强等完成
注释来源:
[1] CounterPoint Research. Global Smartphone OS Market Share Q4 2024. 2024. Web.
https://www.counterpointresearch.com/
[2] The Linux Foundation. "The Linux Foundation: It's Not Just the Linux Operating System." Linux Foundation, 21 Jan. 2020,
www.linuxfoundation.org/blog/blog/the-linux-foundation-its-not-just-the-linux-operating-system.
[3] Yang, Zeyi. "How Chinese AI Startup DeepSeek Made a Model that Rivals OpenAI." Wired, 25 Jan. 2025,
https://www.wired.com/story/deepseek-china-model-ai/?utm_source=chatgpt.com.
[4] 中国通信标准化协会. 全球开源生态洞察报告(2024 年)[EB/OL]. (2024-07-05) [2025-03-25].
http://221.179.172.81/images/20240705/27981720160008486.pdf.
[5] David, Emilia. "Meta Says Llama 3 Beats Most Other Models, Including Gemini." The Verge, 18 Apr. 2024,
https://www.theverge.com/2024/4/18/24134103/llama-3-benchmark-testing-ai-gemma-gemini-mistral.
[6] Raymond, Eric S. The Cathedral and the Bazaar: Musings on Linux and Open Source by an Accidental Revolutionary. O'Reilly Media, 1999.
[7] Third Way. "Open-Source AI Is a National Security Imperative." Third Way, February 2025, www.thirdway.org/report/open-source-ai-is-a-national-security-imperative. Accessed 25 Mar. 2025.
[8] National Telecommunications and Information Administration. "Dual-Use Foundation Models with Widely Available Model Weights." National Telecommunications and Information Administration, July 2024, .
[9] Considerations for Governing Open Foundation Models,
https://hai.stanford.edu/sites/default/files/2023-12/Governing-Open-Foundation-Models.pdf
[10] Raji, Timnit. "Godmother of AI Says California's AI Bill Will Harm U.S. Ecosystem." Fortune, 6 Aug. 2024,
https://fortune.com/2024/08/06/godmother-of-ai-says-californias-ai-bill-will-harm-us-ecosystem-tech-politics/?abc123.
[11] "California SB 1047: Executive's Guide." FairNow, https://fairnow.ai/california-sb-1047-executives-guide/.
[12] Gent, Edd. "Protesters Decry Meta’s 'Irreversible Proliferation' of AI." IEEE Spectrum, 6 Oct. 2023,
https://spectrum.ieee.org/meta-ai#:~:text=The%20self,how%20the%20AI%20is%20used. Accessed 26 Mar. 2025.
[13] La Malfa, Emanuele, et al. "Language-Models-as-a-Service: Overview of a New Paradigm and its Challenges." arXiv,
https://ar5iv.labs.arxiv.org/html/2309.16573v2#:~:text=et%C2%A0al.
[14] Karabus, Jude. "GitHub CEO says EU AI Act shouldn't apply to open source devs." The Register, 7 Feb. 2023,
https://www.theregister.com/2023/02/07/github_ceo_ai_act/.
[15] "AI Act: How the EU Can Address the Challenges Posed by General-Purpose AI Systems." OpenFuture, Jan. 2023,
https://openfuture.eu/wp-content/uploads/2023/01/AI-Act_Mozilla-GPAI-Brief_Kx1ktuk.pdf.
[16] Denton, Jake. "The U.S. Shouldn't Go the Way of Europe on AI." The Heritage Foundation, 8 May 2024,
https://www.heritage.org/big-tech/commentary/the-us-shouldnt-go-the-way-europe-ai.
[17] Solaiman, Irene, 等. "Ethics and Society Newsletter #3: Ethical Openness at Hugging Face." Hugging Face Blog, 30 Mar. 2023,
https://huggingface.co/blog/ethics-soc-3#safeguards.
[18] Castaño, Joel, 等. "Analyzing the Evolution and Maintenance of ML Models on Hugging Face." arXiv, 版本2, 2024年2月5日, https://arxiv.org/pdf/2311.13380.
[19] U.S. AI Safety Institute. "Managing Misuse Risk for Dual-Use Foundation Models." National Institute of Standards and Technology, 2025年1月,