Post

It’s useful to think of frontier labs as a protection racket. Many use cases can settle for “good enough” models from other vendors. But cyber defense alone is worth spending a decent % of GDP on a protective swarm of the absolute best models.

16/10/26, 1:19 am0 reposts0 likes

Dean Lee

@deanlee.info

In an adversarial arms race, the payoff function is strictly convex: a second-best model provides zero marginal defense against a frontier exploit. That shifts lab pricing from enterprise productivity software to sovereign deterrence, where buyer price elasticity disappears at the tail.