I fed it some of my old HN comments, comfortably longer than the 1000 character minimum and found 9/10 times the classifier marked it as "unclear". This was a c

I fed it some of my old HN comments, comfortably longer than the 1000 character minimum and found 9/10 times the classifier marked it as "unclear". This was a comment from 2020. I found out that just repeating a sentence a few times causes it to classify something as "likely". This is not only an unwinnable race, I know for a fact some teachers will use anything above unlikely to mean used AI. At some point in the future, compute will be cheap enough to where a lot of online content will be put through a similar classifier. I am curious to know how conservative the estimates are. This was a non-technical comment, I wonder if a more technical comment would be even more likely to be misclassified.