If you can parallel derive something from other sources, it's not that private.

This is the intuitive logic behind k-anonymity.

But it also just makes sense: if multiple independent users make some kind of decision, then the more people that do it, the less private the information is.

If there's an action that you know has high intention, then if even a handful of users do it, that's a very good signal that it's high-intent and potentially generalizable to other users.

If lots of users have a given need and seem to not have a good answer to it, that's the adjacent possible, the frontier of user demand.

You can imagine publishing a dashboard to help creators in the ecosystem know what kinds of use cases to invest time in building.

A machine could help sort through the adjacent possible, allowing the swarm of creators to fill it in more efficiently.