Subvolts take Understand how advanced AI models like GPT-5 actually reason and learn from an OpenAI insider. Oct 16, 2025 - Developers
- Learn about the practical implementation and trade-offs behind advanced AI reasoning and tool use.
- Business
- Understand the core technologies driving AI progress and how labs like OpenAI are organized for innovation.
- Everyday users
- Gain insight into how AI models like GPT-5 actually work and what their "thinking" process involves.
Features a top OpenAI researcher explaining the inner workings of a highly anticipated model, GPT-5, offering direct insight into AI's future.
How GPT-5 Thinks — OpenAI VP of Research Jerry Tworek
Audio streams from Matt Turck. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
OpenAI VP of Research Jerry Tworek discusses how GPT-5 "thinks" on The MAD Podcast. He explains the roles of pretraining and reinforcement learning, the function of on-screen thinking, and when extra compute helps. The conversation traces model evolution from O1 to GPT-5, covering verifiers, reward design, and auto reasoning trade-offs. Tworek also details OpenAI's research organization and transparency.
This episode explains how AI like GPT-5 "thinks" by combining training data with rewards. It covers why this process helps AI solve complex problems and how OpenAI builds these advanced systems.
This episode offers a deep dive into the practical workings of modern AI reasoning models, directly from an OpenAI leader. It's for anyone curious about how systems like GPT-5 achieve their capabilities, beyond the surface level. The discussion on pretraining, RLHF, and the evolution of models connects directly to the current frontier research efforts by major labs like Google Gemini and Anthropic, highlighting the core technical challenges and approaches.
- HostMatt Turck
- GuestJerry TworekVP of Research, OpenAI
- Guest: Jerry Tworek, OpenAI VP of Research
- Title: How GPT-5 Thinks
- Show: The MAD Podcast
- Length: 76 minutes
- Model Evolution Mentioned: O1, O3, GPT-5 (03.1-ish)
What does it really mean when GPT-5 “thinks”? In this conversation, OpenAI’s VP of Research Jerry Tworek explains how modern reasoning models work in practice—why pretraining and reinforcement learning (RL/RLHF) are both essential, what that on-screen “thinking” actually does, and when extra test-time compute helps (or doesn’t). We trace the evolution from O1 (a tech demo good at puzzles) to O3 (the tool-use shift) to GPT-5 (Jerry calls it “03.1-ish”), and talk through verifiers, reward design, and the real trade-offs behind “auto” reasoning modes.
We also go inside OpenAI: how research is organized, why collaboration is unusually transparent, and how the company ships fast without losing rigor. Jerry shares the backstory on competitive-programming results like ICPC, what they signal (and what they don’t), and where agents and tool use are genuinely useful today. Finally, we zoom out: could pretraining + RL be the path to AGI?
This is the MAD Podcast —AI for the 99%. If you’re curious about how these systems actually work (without needing a PhD), this episode is your map to the current AI frontier.
OpenAI
Website - https://openai.com
X/Twitter - https://x.com/OpenAI
Jerry Tworek
LinkedIn - https://www.linkedin.com/in/jerry-tworek-b5b9aa56
X/Twitter - https://x.com/millionint
FIRSTMARK
Website - https://firstmark.com
X/Twitter - https://twitter.com/FirstMarkCap
Matt Turck (Managing Director)
LinkedIn - https://www.linkedin.com/in/turck/
X/Twitter - https://twitter.com/mattturck
(00:00) Intro
(01:01) What Reasoning Actually Means in AI
(02:32) Chain of Thought: Models Thinking in Words
(05:25) How Models Decide Thinking Time
(07:24) Evolution from O1 to O3 to GPT-5
(11:00) Before OpenAI: Growing up in Poland, Dropping out of School, Trading
(20:32) Working on Robotics and Rubik's Cube Solving
(23:02) A Day in the Life: Talking to Researchers
(24:06) How Research Priorities Are Determined
(26:53) Collaboration vs IP Protection at OpenAI
(29:32) Shipping Fast While Doing Deep Research
(31:52) Using OpenAI's Own Tools Daily
(32:43) Pre-Training Plus RL: The Modern AI Stack
(35:10) Reinforcement Learning 101: Training Dogs
(40:17) The Evolution of Deep Reinforcement Learning
(42:09) When GPT-4 Seemed Underwhelming at First
(45:39) How RLHF Made GPT-4 Actually Useful
(48:02) Unsupervised vs Supervised Learning
(49:59) GRPO and How DeepSeek Accelerated US Research
(53:05) What It Takes to Scale Reinforcement Learning
(55:36) Agentic AI and Long-Horizon Thinking
(59:19) Alignment as an RL Problem
(1:01:11) Winning ICPC World Finals Without Specific Training
(1:05:53) Applying RL Beyond Math and Coding
(1:09:15) The Path from Here to AGI
(1:12:23) Pure RL vs Language Models
- pretraining
- Jerry Tworek
- Initial training of AI models on vast amounts of data.
- reinforcement learning (RL/RLHF)
- Jerry Tworek
- Training AI by rewarding desired behaviors, often with human feedback.
- verifiers
- Jerry Tworek
- Systems that check if an AI's output is correct or meets certain standards.
- auto reasoning modes
- Jerry Tworek
- AI features that automatically attempt to reason through problems.
Same guests, other shows
49 minBuilding the Automated AGI Lab: Core Automation's Jerry Tworek and Rohan Anil
WithJerry Tworek, Rohan Anil
1 h 03Samsung Galaxy XR Gets Real
WithTom Merritt
1 h 18Why GM will give you Gemini — but not CarPlay
WithMary Barra, Sterling Anderson
1 hGoogle Researcher Shows Life "Emerges From Code" - Blaise Agüera y Arcas
WithBlaise Agüera y Arcas
44 minQuestions people ask
- What does the "thinking" process in models like GPT-5 entail?
- The on-screen "thinking" refers to how modern reasoning models work in practice, involving pretraining and reinforcement learning to process information and generate responses.
- How is research organized at OpenAI?
- The episode touches on how research is organized at OpenAI, emphasizing unusually transparent collaboration and fast, rigorous shipping practices.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Matt Turck.