This episode with Jakub Pachocki (Chief Scientist, OpenAI) is a frank conversation about the race to models that don't just talk, but independently conduct research, write programs and discover new theorems. We break down the O1 (free reasoning) breakthrough, the role of reinforcement learning in teaching "courses of thought," the limitations of data and the perspective of models that train even better models. Kuba also talks about long-term memory, robotics, the "internal monologue" of models, the pace of change from GPT-2 to O1, and finally, very human concerns, work routines and advice for Polish talent.