Core Memory OpenAI's Research Chief On The Soup Wars, Poker And The Next Models - EP 46 Mark Chen
338 snips
Dec 1, 2025 
Mark Chen, Chief Research Officer at OpenAI, discusses the challenges of prioritizing research in a fast-paced AI landscape. He shares insights on the intense recruiting competition with Meta, including Zuckerberg’s unique soup deliveries to attract top talent. The conversation also delves into future AI models, particularly in response to Google's Gemini 3, and highlights Chen's competitive nature honed through coding contests. Additionally, he explains the critical focus on pre-training and how OpenAI aims to democratize scientific discovery.
AI Snips
Chapters
Transcript
Episode notes
AI Brings New Intuition To Hard Problems
- Models approach contest problems differently from humans, sometimes finding tasks easier that humans find ad hoc.
- Chen sees human-AI collaboration unlocking new frontier research because AI has different intuitions about difficulty.
GPT-5 Pro Aha Moment In Physics
- Chen recalls a physicist who fed a paper to GPT-5 Pro and saw the model 'think for 30 minutes and just get it.'
- He compares that moment to AlphaGo's surprising moves and predicts more frontier science breakthroughs.
Pre-Training Muscle And Upside Remain
- OpenAI rebuilt its pre-training 'muscle' after focusing on reasoning for years and now sees more room in scaling pre-training.
- Chen argues scaling and algorithmic improvements in pre-training remain powerful levers.