Deep nets can't run long programs like Solomonoff induction allows
Kenku_Allaryi · x · 2026-10-10
A technical exchange about Solomonoff induction vs deep learning: Solomonoff's framework permits programs with short code but very long runtimes, while deep networks have bounded running time and thus cannot do this. The reply draws an analogy to humans, who also exhibit a deep-net-like inductive bias — our local weights likely solve calculus problems latently all the time, yet it took a genius like Newton to see calculus explicitly.