Greg Brockman on OpenAI's Road to AGI

In this episode of the Latent Space Podcast, Alessio and swyx interview Greg Brockman from OpenAI about the recent releases of GPT-5 and GPT-OSS. Brockman discusses the development of reasoning in AI models, the shift towards online learning, and the importance of compute in AI advancement. The conversation covers the capabilities of GPT-5, its applications in various domains like mathematics and coding, and the potential for personalized AI experiences. They also explore the engineering challenges at OpenAI, the value of engineers in the age of AI, and the future of AI research, including the role of open source models and the balance between AI capabilities and safety.

Outlines

Sign in to continue reading, translating and more.

Continue

Latent Space: The AI Engineer Podcast

Introduction to GPT-5 and the Evolution of Reasoning in AI

Online Learning, Compute, and the Role of Human Data Curators in AI

Compute as a Driver of Intelligence and the Transferability of Learning

Addressing Time Barriers in RL and the Biological Basis of Neural Nets

The GPT-5 Era: Intelligence and Intellectual Feats

Evaluating Model Performance and Extracting Potential Energy

AI Productization and Agent Robustness

Psychohistory of LLMs and the Future of Model Preferences

Generalization in Deep Learning and the Router in GPT-5

Pricing, Self-Improving Agents, and Architectural Decisions in GPT-OSS

Future Architectures and the Importance of American Open Source

Engineering at OpenAI and the Value of Engineers

The State of AI Research and Prioritization at OpenAI

Advice for the Future and the Abundance of Problems

Greg Brockman on OpenAI's Road to AGI

Latent Space: The AI Engineer Podcast

00:04Introduction to GPT-5 and the Evolution of Reasoning in AI

Introduction to GPT-5 and the Evolution of Reasoning in AI

04:02Online Learning, Compute, and the Role of Human Data Curators in AI

Online Learning, Compute, and the Role of Human Data Curators in AI

07:46Compute as a Driver of Intelligence and the Transferability of Learning

Compute as a Driver of Intelligence and the Transferability of Learning

13:21Addressing Time Barriers in RL and the Biological Basis of Neural Nets

Addressing Time Barriers in RL and the Biological Basis of Neural Nets

18:49The GPT-5 Era: Intelligence and Intellectual Feats

The GPT-5 Era: Intelligence and Intellectual Feats

22:51Evaluating Model Performance and Extracting Potential Energy

Evaluating Model Performance and Extracting Potential Energy

27:30AI Productization and Agent Robustness

AI Productization and Agent Robustness

32:55Psychohistory of LLMs and the Future of Model Preferences

Psychohistory of LLMs and the Future of Model Preferences

37:21Generalization in Deep Learning and the Router in GPT-5

Generalization in Deep Learning and the Router in GPT-5

42:21Pricing, Self-Improving Agents, and Architectural Decisions in GPT-OSS

Pricing, Self-Improving Agents, and Architectural Decisions in GPT-OSS

47:53Future Architectures and the Importance of American Open Source

Future Architectures and the Importance of American Open Source

51:32Engineering at OpenAI and the Value of Engineers

Engineering at OpenAI and the Value of Engineers

58:21The State of AI Research and Prioritization at OpenAI

The State of AI Research and Prioritization at OpenAI

1:03:05Advice for the Future and the Abundance of Problems

Advice for the Future and the Abundance of Problems