Subvolts take OpenAI's Yann Dubois explains how AI is becoming reliable and useful, not just capable. May 21, 2026 - Developers
- Learn about the post-training techniques and evaluation methods making frontier models more reliable for real-world applications.
- Business
- Understand how AI is moving from raw capability to dependable systems, impacting enterprise adoption and future product development.
This episode features a leader from OpenAI discussing the practical advancements and future directions of their frontier models.
OpenAI's Yann Dubois: Why AI Progress Suddenly Feels Real
Audio streams from Matt Turck. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
OpenAI's Yann Dubois discusses why AI progress feels significant, focusing on the shift from raw capability to reliable systems. He explains the post-training behind models like GPT-5.5, the role of reinforcement learning in real-world applications, and the importance of continual learning and evals in frontier models. The episode covers how reasoning models work and the differences between GPT-5.5 Thinking and Pro.
AI models are suddenly feeling much more useful, like crossing a big milestone. This episode explains how OpenAI makes its advanced AI, like GPT-5.5, work reliably for real-world tasks, not just puzzles. It's about making AI dependable for everyone.
This episode offers insight into OpenAI's current focus on making AI models more useful and dependable, moving beyond pure capability. It's for AI practitioners and enthusiasts interested in the practical challenges of deploying advanced models. The discussion on post-training, reasoning, and continual learning directly relates to the ongoing efforts by all frontier labs to improve model reliability and real-world performance, a key area of competition.
- HostMatt Turck
- GuestYann Duboisco-lead of the Post-training Frontiers team, OpenAI
- Guest: Yann Dubois, co-lead of the Post-training Frontiers team at OpenAI
- Show: The MAD Podcast
- Length: 74 minutes
- Model Mentioned: GPT-5.5
- Key Focus: Shift from raw model capability to useful, reliable systems
AI suddenly feels like it has crossed a threshold, and Yann Dubois, co-lead of the Post-training Frontiers team at OpenAI, joins Matt Turck to explain why. Yann’s team has led the post-training behind the company's reasoning models, including the recent GPT-5.5 release. In this conversation, we go inside the shift from raw model capability to useful, reliable systems: what changed with GPT-5.5, why reinforcement learning is moving beyond math and coding competitions into messy real-world work, how reasoning models like GPT-5.5 actually work, the difference between GPT-5.5 Thinking and GPT-5.5 Pro, why post-training has become one of the most important frontiers in AI, and why evals, model-as-judge, hallucinations, agentic workflows, GDPval, and continual learning are now central to the next phase of frontier models. Yann also shares why continual learning remains one of AI's biggest unsolved problems three years after ChatGPT, and where startups still have massive room to build as frontier models race ahead.
(00:00) - Cold open
(00:34) - Intro
(01:30) - Why recent AI progress feels like a step function
(04:13) - Model reliability & the rollercoaster of shipping 5.5
(07:33) - How OpenAI structures vertical and horizontal teams
(09:49) - Improving model efficiency and test-time compute
(12:32) - Yann Dubois' journey from Switzerland to OpenAI
(15:37) - Reasoning in 2026: Real-world utility vs verifiable rewards
(18:34) - GPT-5.5 Thinking vs Pro: Scaling test-time compute
(20:09) - How reasoning models become more efficient
(23:23) - Pre-training scaling and overcoming the data wall
(27:03) - Multimodal data, synthetic data, and embodied AI
(31:05) - Demystifying mid-training and post-training
(37:21) - Does RL create new capabilities in AI?
(38:53) - The challenges and frontier of scaling RL
(43:09) - Is building AI models a craft or a strict science?
(48:21) - How AI models generalize across different domains
(54:18) - How reinforcement learning cures AI hallucinations
(56:04) - Negative generalization and conflicting instructions
(58:05) - Can RL scale to law, medicine, and the broader economy?
(1:00:19) - The evaluation bottleneck and Model as a Judge
(1:04:21) - Continuous AI progress & continual learning
(1:08:49) - Will foundation models eat the agent harness?
(1:11:23) - Why startups should focus on the last mile of AI
- post-training
- Yann Dubois
- The process of refining AI models after their initial training to improve specific behaviors or capabilities.
- reinforcement learning
- Yann Dubois
- A type of machine learning where an AI learns by trial and error, receiving rewards for good actions and penalties for bad ones.
- evals
- Yann Dubois
- Short for evaluations, these are tests used to measure an AI model's performance and capabilities.
- hallucinations
- Yann Dubois
- When an AI generates incorrect or nonsensical information that is presented as fact.
41 min
57 min
Questions people ask
- What is the difference between GPT-5.5 Thinking and GPT-5.5 Pro?
- The notes mention this difference relates to scaling test-time compute, but do not provide further detail.
- What are some key areas for future AI development?
- Continual learning remains a major unsolved problem, and startups have significant opportunities in areas like evals, model-as-judge, and agentic workflows.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Matt Turck.