"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis Chinese AI – They're Just Like Us? With Beijing-Based Concordia AI CEO Brian Tse
66 snips
Oct 18, 2025 Brian Tse, the Beijing-based founder and CEO of Concordia AI, dives into China's unique approach to AI safety and governance. He discusses the nation's proactive testing regulations and public optimism about AI's role in economic growth. With insights on the interplay between capabilities and safety, he highlights China's advancements in embodied AI and humanoid robots. Tse also emphasizes the importance of international cooperation and shared responsibility in AI development, addressing concerns over labor displacement and the need for standardized frameworks.
AI Snips
Chapters
Transcript
Episode notes
Standards Address Catastrophic Risks
- China's national standards (TC260) explicitly address catastrophic risks including loss of control, misuse, and biological threats.
- Concordia co-authored a Frontier AI Risk Management Framework with Shanghai AI Lab to set red and yellow lines.
DeepSeek R1: Peer Review And Iteration
- DeepSeek R1 was peer-reviewed in Nature and acknowledged vulnerabilities, especially that open models can be fine-tuned to undermine safety.
- Concordia's tests showed DeepSeek improved biological-task refusals from ~11% to ~55% across versions.
Open Weights With Pre-Training Controls
- China supports open weights but recognizes misuse risks and is exploring pre-training filtering to remove hazardous knowledge.
- Policymakers and standards bodies are considering controls that iterate as new threats emerge.