开放模型：未来十年 AI 研究的引擎

Open Models as the Engine for the Next Decade of AI Research

内森·兰伯特 Nathan Lambert · Turing Post · 2026-02-06 · 约 47 分钟 · 原视频 ↗

打开互动全文版（中英对照 + 朗读 + 问答）→

本期速览 · Overview

Nathan Lambert 探讨开放模型在 AI 研究、地缘政治中的作用，以及影响力从美国向中国的转移。

Nathan Lambert discusses the role of open models in AI research, geopolitics, and the shift of influence from the US to China.

要点 · TL;DR

• 开放模型将驱动未来十年的 AI 研究，促进学术探索。
  Open models will drive AI research for the next decade, enabling academic exploration.
• 预训练数据因版权问题是最难开放的合法部分。
  Pre-training data is the hardest legal part to open due to copyright issues.
• 最佳开放模型落后前沿闭源模型 6-9 个月，这一差距可能持续。
  Best open models lag 6-9 months behind closed frontier models, a gap that may persist.

核心观点 · Key points

• 开放模型将成为未来 10 年 AI 研究的引擎，推动学术探索。
  Open models will be the engine for AI research for the next 10 years, enabling academic exploration.
• 预训练数据因版权问题是最难合法开放的环节。
  Pre-training data is the hardest legal part to get open due to copyright issues.
• 最佳开放模型落后前沿闭源模型 6-9 个月，这一差距可能持续。
  The best open models lag 6-9 months behind closed frontier models, and this gap may persist.
• 基于强化学习的后训练是工具使用和编码智能体的关键，但复杂度高。
  Post-training with reinforcement learning is key for tool use and coding agents, but complex.
• 中国开放模型受意识形态驱动，已形成开放行业标准。
  Chinese open models are ideologically driven and have created an industry standard for openness.
• 开放模型对主权 AI 和减少权力集中至关重要。
  Open models are crucial for sovereign AI and reducing concentration of power.

反共识 · Contrarian takes

• 开放模型被过度炒作，实际使用中不如闭源模型。
  Open models are overhyped; they are not as good as closed models for real-world use.
• 机器人技术的时间线似乎过早，现在规模化机器人可能为时过早。
  The timeline on robotics seems too soon; scaling robotics now may be premature.
• 马斯克的企业有反派气质，但他的过往记录表明有战略规划。
  Musk Industries has villain vibes, but his track record suggests a strategic plan.
• 学术界影响力低迷，开放模型可重振其在 AI 研究中的作用。
  Academia is in a lull of influence; open models can revive its role in AI research.
• AI 前沿正从基准分数转向产品体验和智能体系统。
  The frontier of AI is shifting from benchmark numbers to product experience and agentic systems.
• 由于资源和人才差距，开放模型可能无法赶上闭源模型。
  Open models may not catch up to closed models due to resource and talent gaps.

本期章节 · Chapters（共 14）

• 引言与 Nathan 的崛起 Introduction and Nathan's Rise to Prominence
• 开放模型为何重要 Why Open Models Matter Despite Lower Performance
• 中国开放模型与地缘政治 Chinese Open Models and Geopolitics
• 中美开放模型生态对比 China vs US open model ecosystem
• 开放生态的研究转变 Research shifts in open ecosystem
• 开放模型训练的挑战 Challenges in open model training
• 开放与封闭模型的差距 Open vs Closed Models Gap
• AI 研究与混合模型时间线 Timeline of AI research and hybrid models
• 开放模型与许可 Open Models and Licensing
• 模型开发中的小众问题平衡 Balancing Niche Problems in Model Development
• 未来方向：智能体与超越 Future Directions: Agents and Beyond
• 资源挑战与开源 Resource challenges and open source
• 开放模型与 AGI Open Models and AGI
• 书籍推荐 Book Recommendation

阅读全文双语转录 →