It definitely hasn’t worked then. How’s that approach any efficient?

When many people make the claim “it hasn’t worked” wrt some intervention they often miss the point: the key question is change relative to the counterfactual; i.e. having done nothing.

So, to evaluate action at t=0, we compare some metric at the real t=1 against the counterfactual at t=1.

It is logically invalid to evaluate the efficacy of an action by only comparing the metric at t=0 and t=1. That kind of reasoning error is incredibly common.

Indexing people's online activity (in order to search for specific kinds of activiy) is a far cry from watching everything everybody is doing, let alone acting on all that information.

Wait until LLMs meet that data