Tetragrammaton with Rick Rubin Greg Brockman (Part 2)
280 snips
Feb 28, 2026 
Greg Brockman, OpenAI cofounder and president who helped build ChatGPT and Codex, walks through AI milestones and the switch from rule-based NLP to deep learning. He recounts founding OpenAI, scaling research into products, agentic AI that connects models to tools, and the engineering and safety trade-offs of deploying powerful models.
AI Snips
Chapters
Transcript
Episode notes
Short Iteration Loops Beat Line-By-Line Reviews
- To accelerate researcher–engineer loops, work tightly in short iterations: propose several ideas, have the researcher eliminate bad ones, and use feedback to converge quickly.
- Greg applied this as an engineer at OpenAI to avoid slow line-by-line reviews and speed progress.
OpenAI's Three Step Plan Guided A Decade Of Work
- At a 2015 offsite the founding team wrote a three-step plan: solve reinforcement learning, solve unsupervised learning, then scale to more complex tasks.
- Greg says OpenAI's decade of work is a growth of that original tightly held vision.
Vibe Coding Turns Engineers Into Managers Of Agents
- Vibe coding with models moves the machine closer to humans, shifting engineers toward managerial roles supervising agents rather than typing every line.
- Greg observed a demo where Codex built a JavaScript game live and later generated months-old hand-coded sites in minutes.