Mind Over Models: Bob McGrew on Scaling Reasoning Toward AGI

This episode features former OpenAI researcher Bob McGrew tracing AI’s shift from data-heavy pretraining to a new era driven by reasoning and test-time compute. He explains how scaling reasoning capabilities (rather than just model size or datasets) could produce reliably autonomous agents that perform complex tasks for users. McGrew also addresses the remaining challenges in grounding AI in the physical world, cultural changes in research, and a future where humans act as “geniuses” and “managers” of automated systems. Tune in for a clear view of the technical and social road toward AGI.