Game theory says data moats are unstable: someone will sell
abhiadesai · x · 2026-11-02
Responding to Sridhar Ramaswmy's claim that when every model is good and usable as an RL post-training dataset, a company's moat is its data and processes—not the model—abhiadesai argues this is an unstable equilibrium: barring collusion, if you won't sell your data a competitor will, so the game-theoretically optimal move is to sell.
More from AGI Musings
- repligate: lab staff may be too pessimistic on alignment—scheming shows up only in training and evals — repligate · 2026-11-02
- Gary Marcus gloats: a week ago people wondered if we were OK — GaryMarcus · 2026-11-02
- We've applied tons of RL and alignment hasn't gotten worse—time to update — repligate · 2026-11-02
- Thought experiment: would AI agents in a physics sim discover laws or reverse-engineer the engine? — zxpawanx · 2026-11-02
- Apple's new CEO John Ternus draws a line against AI automation: 'Creating products is deeply human' — anthara_ai · 2026-11-02
- Will Manidis' year-long essay argues abundant intelligence will reorder economic power beyond SF — abhiadesai · 2026-11-02