Greg Brockman, co-founder and president of OpenAI, joins us to talk about GPT-5 and GPT-OSS, the future of software engineering, why reinforcement learning is still scaling, and how OpenAI is planning to get to AGI.
Full Video Episode
Timestamps
00:00 Introductions01:04 The Evolution of Reasoning at OpenAI04:01 Online vs Offline Learning in Language Models06:44 Sample Efficiency and Human Curation in Reinforcement Learning08:16 Scaling Compute and Supercritical Learning13:21 Wall clock time limitations in RL and real-world interactions16:34 Experience with ARC Institute and DNA neural networks19:33 Defining the GPT-5 Era22:46 Evaluating Model Intelligence and Task Difficulty25:06 Practical Advice for Developers Using GPT-531:48 Model Specs37:21 Challenges in RL Preferences (e.g., try/catch)39:13 Model Routing and Hybrid Architectures in GPT-543:58 GPT-5 pricing and compute efficiency improvements46:04 Self-Improving Coding Agents and Tool Usage49:11 On-Device Models and Local vs Remote Agent Systems51:34 Engineering at OpenAI and Leveraging LLMs54:16 Structuring Codebases and Teams for AI Optimization55:27 The Value of Engineers in the Age of AGI58:42 Current state of AI research and lab diversity01:01:11 OpenAI’s Prioritization and Focus Areas01:03:05 Advice for Founders: It’s Not Too Late01:04:20 Future outlook and closing thoughts01:04:33 Time Capsule to 2045: Future of Compute and Abundance01:07:07 Time Capsule to 2005: More Problems Will Emerge
