Nathan Lambert: inference-time compute will scale big, but intelligence explosion isn't near

natolambert · x · 2026-09-19

Nathan Lambert publishes a "moderate's view" on recent AI events: he admits underestimating how much inference-time compute will scale soon, with frontier labs like OpenAI and Anthropic now running thousands of concurrent agents. But he argues this differs from RSI and sees insufficient evidence that an intelligence explosion is near. In the fuller post he contends the anxious, hyper-competitive culture of frontier labs amplifies risk narratives, and that exaggerating timelines carries negative second-order effects.

Original post →

More from AGI Musings

AGI Musings channel →