
ChinaTalk Overfit: Shenzhen vs SF Vibes, Model Progress, Jordan a Berkeley Bowl Hater
55 snips
Sep 25, 2025 Jasmine Sun, a technology and China-focused writer, shares vibrant insights from her recent trips to Shenzhen and Shanghai, highlighting the optimism within China's tech scene. Nathan Lambert, an AI analyst, discusses advancements in AI models and coding agents, emphasizing the practical applications of these tools. They explore the cultural contrasts between the Bay Area's vibe and Shenzhen's dynamic energy. The conversation also touches on the interplay between Western and Chinese tech ecosystems and the intriguing challenges of AI development.
AI Snips
Chapters
Books
Transcript
Episode notes
Try Coding Agents With A Small Project
- Try coding agents (Copilot, Cursor, ChatGPT Codex) by building small personal projects like websites to learn quickly.
- Use side-by-side comparisons to see how product interfaces improve the same underlying model.
Start By Building A Personal Website
- If you pay for ChatGPT, enable Codex/CLI and use it to build a personal website or resume to learn tooling quickly.
- Let the agent handle hosting steps and integrations to remove technical barriers and ship a usable site fast.
Use A Personal Eval To Track Models
- Develop a personal eval task (e.g., Jordan's book-highlight reviews) to compare models consistently over time.
- Consistent, repeatable tasks reveal real improvements and product-level differences between models.






