The discovery of some information theoretic limit to model learning (at least using current paradigms) can't come soon enough.