From Vibe Coding to Vibe Researching: OpenAI’s Mark Chen and Jakub Pachocki

In this episode of the a16z Podcast, Anjney Midha and Sarah Wang interview OpenAI's Chief Scientist Jakub Pachocki and Chief Research Officer Mark Chen about GPT-5, the company's advancements in AI reasoning, and their pursuit of an automated researcher. They discuss the shift towards economically relevant benchmarks for AI evaluation, the continuous improvements from reinforcement learning (RL), and the development of new codecs for real-world coding. The conversation explores the balance between fundamental research and product development, the importance of fostering a creative and resilient research culture, and the ongoing significance of compute in AI progress.

Outlines

Part 1: GPT-5 Introduction and Evaluation

Part 2: Reinforcement Learning and Coding

Part 3: Qualities of Researchers and Research Organization

Part 4: Balancing Research and Product Development

Part 5: Long-Term Vision and Collaboration

Sign in to continue reading, translating and more.

Continue

The a16z Show

Part 1: GPT-5 Introduction and Evaluation

Introduction to GPT-5 and the Shift Towards Reasoning

Surprising Capabilities of GPT-5 and the Research Roadmap

Agency, Reasoning, and Verifiable Domains

Part 2: Reinforcement Learning and Coding

The Continued Success of Reinforcement Learning (RL)

GPT-5 Codecs and the Evolution of Coding

AI's Impact on Coding and the Future of Research

Part 3: Qualities of Researchers and Research Organization

Qualities of a Great Researcher: Persistence, Honesty, and Experience

Problem Selection and Maintaining a Resilient Research Organization

Hiring Researchers and Fostering a Creative Culture

Part 4: Balancing Research and Product Development

Balancing Fundamental Research and Product Development

Prioritization and Resource Allocation in Research

Resource Allocation: Compute, Data, and People

Bridging the Gap Between Academia and Frontier AI

Part 5: Long-Term Vision and Collaboration

Long-Term Vision and Constant Change

The Importance of Trust and Collaboration

From Vibe Coding to Vibe Researching: OpenAI’s Mark Chen and Jakub Pachocki

The a16z Show

Part 1: GPT-5 Introduction and Evaluation

00:00Introduction to GPT-5 and the Shift Towards Reasoning

Introduction to GPT-5 and the Shift Towards Reasoning

04:46Surprising Capabilities of GPT-5 and the Research Roadmap

Surprising Capabilities of GPT-5 and the Research Roadmap

07:56Agency, Reasoning, and Verifiable Domains

Agency, Reasoning, and Verifiable Domains

Part 2: Reinforcement Learning and Coding

11:17The Continued Success of Reinforcement Learning (RL)

The Continued Success of Reinforcement Learning (RL)

14:21GPT-5 Codecs and the Evolution of Coding

GPT-5 Codecs and the Evolution of Coding

17:20AI's Impact on Coding and the Future of Research

AI's Impact on Coding and the Future of Research

Part 3: Qualities of Researchers and Research Organization

20:33Qualities of a Great Researcher: Persistence, Honesty, and Experience

Qualities of a Great Researcher: Persistence, Honesty, and Experience

23:33Problem Selection and Maintaining a Resilient Research Organization

Problem Selection and Maintaining a Resilient Research Organization

27:33Hiring Researchers and Fostering a Creative Culture

Hiring Researchers and Fostering a Creative Culture

Part 4: Balancing Research and Product Development

30:38Balancing Fundamental Research and Product Development

Balancing Fundamental Research and Product Development

33:38Prioritization and Resource Allocation in Research

Prioritization and Resource Allocation in Research

37:01Resource Allocation: Compute, Data, and People

Resource Allocation: Compute, Data, and People

40:11Bridging the Gap Between Academia and Frontier AI

Bridging the Gap Between Academia and Frontier AI

Part 5: Long-Term Vision and Collaboration

43:30Long-Term Vision and Constant Change

Long-Term Vision and Constant Change

47:17The Importance of Trust and Collaboration

The Importance of Trust and Collaboration