Cerebras Big Chip Club: Jeffrey Wang (OpenAI) on Turning Compute Into Intelligence

How do OpenAI researchers turn compute into intelligence, and what changes when inference gets dramatically faster? Jeffrey Wang from OpenAI discusses the relationship between pre-training and reinforcement learning, why predictable scaling matters, how researchers co-design models with hardware, and what low-latency inference unlocks for agents and research workflows. 1:51 Why OpenAI? 3:51 What is pre-training and reinforcement learning? 7:58 Turning compute into intelligence 8:46 Choosing the next model 11:59 OpenAI’s 2026 research focus 14:32 Future model capabilities 15:55 The hardware lottery 17:12 Measuring training efficiency 18:56 AI accelerators and hardware choice 19:53 What ultra-fast inference changes 20:31 Long-horizon agents and research 22:13 Co-designing training and inference 23:35 AI and quantitative trading Watch more conversations from Cerebras Big Chip Club and subscribe for future episodes. #AI #MachineLearning #Cerebras