AI training on its own output is (probably) fine if it’s validated, like a lean-verified proof to Navier-Stokes.
Because it being validated as correct resolves the main issue with incestuous training, which is compounding error.
Because it being validated as correct resolves the main issue with incestuous training, which is compounding error.
AI creates novel discovery> incorporates this information > makes new discovery
This is how it works for humans too.
"AI slop", for example, appears a regression toward some "mean".
The same is true for science and Engineering.
What matters is if you have a method of sorting through the repetitive trash
Rather, that AI is unlikely to produce a fresh Marcin Patrzalek[1].