In this case I spot checked a few random strings from it.
For my personal browsing AI prompt-library-tool I use, it has the ability to click on a highlight, and scroll to that corresponding paragraph. It fails to work if there are any errors (although usually the errors are just “slightly different punctuation”), so it’s actually pretty easy to click through and verify.
But if I were building this into an reliably tool I wanted to use at scale, I’d follow it up with a dumb script that checks if the entire paragraphs match, and if not, if there are random subparts that match a given paragraph from the original content, and then reconcile them. (the sort of thing I’m imagining here is a thing that generates interesting highlights from LessWrong posts and some scaffolding for figuring out if you need to read prerequisites)
In this case I spot checked a few random strings from it.
For my personal browsing AI prompt-library-tool I use, it has the ability to click on a highlight, and scroll to that corresponding paragraph. It fails to work if there are any errors (although usually the errors are just “slightly different punctuation”), so it’s actually pretty easy to click through and verify.
But if I were building this into an reliably tool I wanted to use at scale, I’d follow it up with a dumb script that checks if the entire paragraphs match, and if not, if there are random subparts that match a given paragraph from the original content, and then reconcile them. (the sort of thing I’m imagining here is a thing that generates interesting highlights from LessWrong posts and some scaffolding for figuring out if you need to read prerequisites)