Comment by echelon

Oh sure. But that was a different tone and audience.

The "it's stealing" arguments have quieted down. Especially since there are weights trained on fully licensed materials by Adobe and others.

Now that code models can do it, it's a moot point. Data can be found anywhere, and models are pretty good at generalizing out of domain, not unlike human brains.

Engineers finding out these models are good is drawing a lot of the same "slop" / "clanker" arguments that non-artists have been using. These arguments are much less interesting than arguments about copyright and control.