Manifold The Global AI Race: Z.ai and the View from Beijing — #96
69 snips
Oct 9, 2025 Zixuan Li, Director of Product and genAI Strategy at Z.ai, shares insights from Beijing's AI startup scene. He dives into Z.ai's evolution and open-source models, contrasting them with Western AI developments. Discussions reveal the competitive landscape, talent market dynamics, and government support in China. Li highlights the pragmatic view on AGI, emphasizing practical applications rather than hype. He also touches on the challenges of commercialization and the vibrant community pushing innovation, showcasing Beijing's energetic tech culture.
AI Snips
Chapters
Transcript
Episode notes
From Law Student To AI Product Lead
- Zixuan described his path from law and journalism into AI after finding legal tech boring.
- He studied at MIT and CMU, then returned to China to work on product and growth at Z.ai.
Early Ambition And Open-Source Breakthrough
- Z.ai began early with GLM models in 2021 and trained very large parameter models experimentally.
- Open-sourcing GLM-130B in 2023 raised their profile but deep-pocketed rivals later intensified competition.
Mid‑Training For Agentic Capability
- GLM 4.5 aims to be a balanced single model combining reasoning, coding, and agentic tool use.
- Z.ai inserts a 'mid-training' phase with high-quality code, math solutions, and agentic trajectories to build these skills.