Scaling Laws The Open Questions Surrounding Open Source AI with Nathan Lambert and Keegan McBride
Aug 21, 2025
Nathan Lambert, a post-training lead at the Allen Institute for AI, and Keegan McBride, a lecturer at the Oxford Internet Institute, delve into the evolving landscape of open source AI. They discuss the shift towards open-source models and the implications for AI policy and global competition. The conversation highlights challenges in monetization and contrasts the dynamics of open versus closed models, with specific insights on China's advancements. They also address federal funding issues and the critical role of collaboration between government and academia in fostering innovation.
AI Snips
Chapters
Transcript
Episode notes
Openness Is A Gradient
- Openness in AI is a gradient from APIs to full release of weights, code, and data.
- A fully open-source model provides everything needed to recreate it, including datasets.
Why Firms Hold Back
- Companies avoid fully opening stacks due to commercial incentives and liability risks.
- Releasing training datasets increases ease of reproduction and potential copyright exposure.
Open Releases Seed Ecosystems
- Open-source releases build ecosystems that attract developers, lower costs, and extend influence.
- Firms use openness to undercut competitors and seed downstream dependency on their models.