Descriptions:
Latent Space’s cooking series sits down with Mark Chen, Chief Research Officer at OpenAI, for a wide-ranging conversation recorded while preparing Korean tofu stew. Chen traces his path from high-frequency trading to AI research, arguing that traders develop an “unhackable” optimization mindset — working against a brutally honest real-world metric — that transfers directly to research. He notes that OpenAI has deliberately recruited across disciplines including physicists, mathematicians, and traders rather than requiring formal ML PhDs, valuing creative problem-solving over credential patterns.
The conversation explores how OpenAI identifies and develops research talent, with Chen noting that within six to twelve months it becomes clear which researchers have the strongest trajectory — measured by whether their intuitions align with senior researchers and whether they can execute on ideas others haven’t yet recognized. He describes distinct researcher archetypes: those who move fastest on clear opportunities, and those who propose seemingly outlandish moonshot directions that turn out to be transformative. On developing research taste, Chen recommends rigorous replication — carefully reproducing landmark papers like ResNet and PixelCNN to internalize techniques that papers describe only implicitly.
The interview also covers OpenAI’s o1 reasoning model, the role of evals in guiding research direction, scaling law dynamics, and Chen’s broader perspective on the path toward AGI — drawing on his long tenure at OpenAI shaping core research direction from early language models through today’s frontier systems.
📺 Source: Latent Space · Published June 25, 2026
🏷️ Format: Interview
