The Wright Show The Open Source AI Question (Robert Wright & Nathan Lambert)
6 snips
Aug 14, 2025 
In this insightful discussion, Nathan Lambert, an AI researcher at the Allen Institute for Artificial Intelligence, sheds light on the future of open source AI. He tackles the complexities surrounding its definition and implications, especially in light of recent breakthroughs like China’s DeepSeek. Nathan shares his passion for building a robust open source ecosystem in the U.S. and discusses his ATOM Project aimed at supporting American models. The conversation also delves into the ethical dilemmas of AI centralization and the geopolitical rivalry with China.
AI Snips
Chapters
Transcript
Episode notes
What 'Open Source' Means For AI
- Open-source often refers to releasing model weights and training data, not just code.
- Weights are the costly product of pre-training and enable downstream modification and research.
AI2's Llama Post-Training Story
- AI2 took Meta's Llama weights and built open post-training recipes like Tulu.
- They reproduced and improved post-training techniques to share research pipelines with the community.
Open Models Accelerate Research
- A robust open ecosystem lets more researchers explore post-training questions and increases innovation.
- Open models near the frontier shrink the lag between closed labs and public research.