Karin Dalziel on Nostr: I am super curious if the companies also get rid of the scans produced, so they can ...

I am super curious if the companies also get rid of the scans produced, so they can go back and correct the *almost certainly* incorrect OCR they need to feed the AI. I suspect they don't keep them, which means they would still run into the problem of corrupting their model with slop if they are using AI to aid with OCR, with no way to ever fix it.