Moore's law

The talk argues that AI lab revenue could keep growing much faster than compute capacity, putting upward pressure on compute prices. Labs can close that gap through higher margins or higher compute prices, while shifting more capacity to inference is strategically unattractive if they still aim to train better models. As models become more capable, they may earn far more value from each unit of compute, and the most efficient models could command larger premiums; this could make it harder for smaller competitors and price some AI applications out of the market. The speaker doubts compute supply can scale quickly because of limits in chip advances, fab construction, and wafer allocation, while noting that compute could become cheap again in a future era of automated manufacturing.

Bob McGrew traces OpenAI’s shift from early robotics and game projects toward large language models, arguing that scale and generalization connected those efforts. He describes how reasoning and test-time compute may help models overcome pretraining data limits and make agents reliable enough to take actions. He also discusses how AI startups should begin with the strongest available model, then optimize for cost after finding value with users. The talk closes on AI’s effects on work, the need for useful software tailored to people’s workflows, and the prospect of robotics and scientific progress.

The speakers argue that a wider view shows how much life has improved since the 1990s, from access to information and communication to transportation, healthcare, entertainment, and renewable energy. They point to Starlink, self-driving cars, and cheaper space launches as signs that progress is still underway. Technology brings trade-offs, and innovation arrives unevenly, but those drawbacks do not erase its benefits. They encourage listeners to stay optimistic, build solutions, and invest effort in problems they care about.