The Morning Brief DeepSeeking Power: China’s AI Gambit
Jan 31, 2025
Conor Grennan, Chief AI Architect at NYU Stern, and Jeffrey Ding, Assistant Professor at GWU, dive into the emergence of DeepSeek, a disruptive Chinese AI startup shaking up the global tech scene. They discuss how DeepSeek’s cost-effective models threaten giants like NVIDIA and OpenAI. The conversation explores the geopolitical shifts due to U.S. export controls and the implications for global AI leadership, particularly the U.S.-China rivalry. Is DeepSeek a game-changer or just a fleeting trend? Tune in for insights!
AI Snips
Chapters
Transcript
Episode notes
Mixture of Experts
- DeepSeek's AI model uses a "mixture of experts" architecture.
- This differs from large models like OpenAI, which use the entire model for every query.
Reinforcement Learning
- DeepSeek uses reinforcement learning, similar to dog training.
- It learns through feedback, without needing as many human-labeled examples.
Open-Source Advantage
- DeepSeek being open-source allows others to build upon it, unlike proprietary models.
- This fosters collaboration and could accelerate AI development, much like Wikipedia's model.