François CholletVerified account

I don't consider training on millions of games to be unfair, as long as it doesn't memorize those games. DTS is questionable-ish.

The problem is that we "rush" to get a result, AI training should be much wider than just that one problem we want to solve.

You could just end up calling whatever makes them better an "advantage"..

Isn't that the modus operandi of every champion?

We do it this way because is the easy way @BenedictEvans we should use that advantage to build better AI Be better teachers, with AI also

AGI won't need it but sure single purpose AI that knows nothing else? You're looking at the equivalent of an severely autistic genius.

True that human size training dataset is tiny, but only "during his lifetime", but...

... there are instead millions of years of evolution which have shaped this brain so that it quickly learns and adapts to its environment.