OpenAI's Yann Dubois: Why AI Progress Suddenly Feels Real

Yann Dubois, co-lead of the Post-training Frontiers team at OpenAI, joins Matt Turck for an in-depth discussion regarding the recent acceleration in artificial intelligence progress and the shift toward reliable, useful AI systems. The conversation focuses on the changes introduced with models like GPT-5.5, detailing how post-training methods are transforming raw model capability into practical applications. Dubois explains the mechanics of reasoning models and the expansion of reinforcement learning beyond controlled environments into complex, real-world tasks. Specific topics covered include the distinction between GPT-5.5 Thinking and GPT-5.5 Pro, the critical role of post-training in frontier AI development, and the importance of rigorous evaluations, hallucinations mitigation, and agentic workflows.

Furthermore, the discussion addresses central challenges in the next phase of frontier models, such as the ongoing issue of continual learning, which remains largely unsolved three years after the introduction of ChatGPT. Dubois also provides insight into where new opportunities remain for startups despite the rapid advancement of large language models (LLMs) and foundational AI research.