[hhmc]

I think he's specifically alluding to doing something like Bayesian exploration/exploitation -- which would be very interesting. Unfortunately, you'd need to hire at huge scale to assess the outcome of the experiment, and - as he mentions - it would be hard to get hiring departments to accept.

[sharken]

I much prefer the approach also mentioned in that thread.

Here a hiring process was adjusted so that good current employees could actually pass the hiring process.

Of course that doesn't solve the issue on how to get the first few hires.