Anthropic IPO roadshow slide shows superexponential pretraining efficiency gains

minhnguyen · x · 2026-11-14

A slide from Anthropic's IPO roadshow was cited by a departing employee as a major factor in his decision to leave, surprising him that the company made it public. The chart plots pretraining compute efficiency over time on a log scale, showing continued superexponential algorithmic progress.
The author argues Mythos was not a one-off deviation but a point on a superexponential curve: it could already be retrained with a fraction of the original compute.
He notes this doesn't strictly imply a software-only intelligence explosion—the company has been spending more on R&D—but algorithmic improvements don't require extraordinary genius and are on the verge of being automated.

More from AGI Musings