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

Overview

Shownote

Highlights

Transcript

Chapters

Pins

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

The a16z Show

2025/09/25
The a16z Show

The a16z Show

2025/09/25
OverviewShownoteHighlightsTranscriptChaptersPins

Shownote

What comes after vibe coding? Maybe vibe researching. OpenAI’s Chief Scientist, Jakub Pachocki, and Chief Research Officer, Mark Chen, join a16z general partners Anjney Midha and Sarah Wang to go deep on GPT-5—how they fused fast replies with long-horizon...

Highlights

The conversation dives into the next frontier of AI development, focusing on the evolution from basic pattern recognition to systems capable of deep reasoning and autonomous discovery. As models grow more sophisticated, the focus shifts toward building AI that can act as true research partners, driving innovation in science and technology through extended reasoning and real-world application.
01:46
GPT-5 combines instant-response and long-thinking models for better reasoning
07:46
Reasoning is core to long-horizon model operation.
11:18
Language modeling breakthrough allowed RL to be applied to natural language, opening new research directions
13:14
Reward modeling will rapidly evolve toward more human-like learning approaches.
19:42
Coding from scratch feels strange in the age of AI-assisted development
20:14
Researchers must be ready to fail and learn from failures to make progress.
23:54
Identifying bugs in software and thinking is a major breakthrough in research.
26:08
Researchers are hired based on problem-solving ability, not social media visibility.
38:21
Compute is destiny at a research organization like OpenAI
42:16
Robotics will be a major focus in the future of AI
46:01
Trust between two people at OpenAI has remained unchanged through the changes.

Chapters

Introduction & Goals of Automated Researcher
00:00
The Evolution of Reasoning in AI
00:43
Evaluations: From Benchmarks to Real-World Impact
01:46
Surprising Capabilities of GPT-5
05:15
The Research Roadmap: Next 1, 2, 5 Years
06:56
Long-Horizon Agency & Model Memory
07:46
Reasoning in Open-Ended Domains
09:44
The Role and Progress of Reinforcement Learning
11:18
Reward Modeling & Best Practices
13:14
The New Codex: Real-World Coding
14:21
AI vs. Human Coding: The New Default
16:20
What Makes a Great Researcher?
20:07
Persistence, Conviction, and Problem Selection
21:14
Building and Sustaining a Winning Research Culture
26:00
Balancing Product and Fundamental Research
31:45
The Importance of Compute and Physical Constraints
39:00
Maintaining Speed and Learning at Scale
45:50
Trust and Collaboration at OpenAI
47:18

Transcript

Mark Chen: The big thing that we are targeting is producing an automated researcher. So automating the discovery of new ideas, the next set of evals and milestones that we're looking at will involve actual movement on things that are economically relevant....