A dedicated column picker would be far more intuitive: its discoverability is higher, and features like the indeterminate ‘select all’ checkbox and instant search align with the law of approximation.

For our users handling large datasets, this makes for a much faster, more cohesive tool.