I remember that.

That was when they realized the deep learning was largely unnecessary, and they could just use their massive compute resources to brute force the problem space.

Proving that we would greatly benefit from using our compute resources for science rather than showing ads, and then we just kept showing ads.