It also doesn't count the crazy amount of epigenetics that occurs too, and the specialization that becomes emergently interactive with the environment of other cells and outer world.
Still 50Mb is crazy small for what you'd think the genome would be.
It's actually not crazy small considering what you can expect from natural selection (strongly constrained by the bit error rate in reproduction, more weakly constrained by the total number of selection events in the population since some earlier less complex point). And remember it's specifying the learning infant, not the learned human.
(I plan to write more about the likely information content in a reply to another comment.)
Still 50Mb is crazy small for what you'd think the genome would be.