Comment by nl

The labs are spending hundreds of millions of dollars hiring people doing many fairly random (but economically valuable) tasks to collect this tacit knowledge for RL.

It works really well.