@tobygodat order matters. most decision paths are non-commutative, people leave value on the table because they don't consider changes in value & opp. cost over time properly (similar to what you were describing)
@vineettiruvadi assuming all the geometric/invariance priors, would you predict better outcomes for top down or bottom up? or is it nuanced as always and depends
like starting w macro constraints going down to circuit mechanisms vs local symmetries scaling up to larger dynamics
The problem with AI is that itβs causing this massive brain drain where our brightest kids are working on curing cancer and creating abundant energy instead of building derivatives exchanges