@llewelly @darkcisum

> almost 7 years old

I think a difference now compared to 7 years ago is the size of the pile, it's gone from millions, to billions, to trillions of parameters that are all affected by the "training" data in ways that are more inexplicable than ever.