The Scaling Bet and Scaling Laws

A mind-expanding conversation on how models actually work | Emmanuel Ameisen (Anthropic)

Topic 7 of 18

Published Oct 18, 2026 · Clip 24:45–27:41 (2:56)

They discuss the bet that simple objectives plus massive compute would yield intelligence, which OpenAI and later Anthropic's founders made. Emmanuel says the key insight was finding an objective where compute keeps giving returns — compute plus the richness of text. He references the scaling-laws paper by early OpenAI/Anthropic people: pouring in more compute and data predictably improves next-word prediction, an empirical law that also correlates with tasks people care about.