aifaq.wtf

This tweet was about someone who won a book cover illustration contest, was suspected of using AI, denied it, and then the internet went super detective-y on them. Turns out that without a doubt the cover was generated by Midjourney. Really good reading, highly recommended.

While AI detection doesn't really work, the people using the AI are plenty likely to make mistakes that you can catch.

Actual article is here: Instructor Accuses Texas A&M Class of Using ChatGPT, Withholds Grades

“In Grading your last three assignments I have opened my own account for Chat GTP [sic],” the teacher wrote. “I copy and paste your responses in this account and Chat GTP will tell me if the program generated the content. I put everyone's last three assignments through two separate times and if they were both claimed by Chat GTP you received a 0.”

Sigh. The big problem is while this is obviously very very silly, there are plenty of tools that claim to be AI detectors. They don't work.

Language models generate boring text, so plagiarism detectors detect boring, predictable text. Do you know who else generates boring, predictable text? Students writing boring papers about predictable prompts. And journalists.

This isn't LLMs, but stealing content and avoiding being caught has a long, long history. In this situation the paper has been "synonymized," where a selection of words have been replaced with synonyms to evade direct-match detection.