How reliable is Pangram now? In the past, these types of detectors had a really high false positive rate. Is it improved?
When it flags something as high confidence, 3.1 is very reliable. Human-authored text I saw scoring ~70 with 3.0 now scores ~15. With scores above 90, the false positive rate looks to be about zero.
https://www.pangram.com/blog/pangram-3-0-technical
I looked at all the AI signatures and they’re all phrases I use in my very human writing. If you’re anti-AI generated content then stop posting clearly marked AI dross like this I guess.
"work and personal" is an AI signifier?
"'s the thing"?
"across multiple"?
"identity verification" is 11x more likely to appear in AI writing, per the tool?
I call BS.
Ah yes, using AI to verify AI writing, very trustworthy.