Rather than inaccurate "AI detectors" that a) themselves use LLMs and b) have high false positive rates for nonwhite and non-native English speakers, I'd really like to see version control and metadata analysis becoming the next front in detecting quote-unquote #AI writing. Like, don't just turn in the final draft, but also your developmental drafts, your doodles, your notes. Show you did the work beyond copy-paste and a bit of light editing.

I don't even advocate this method because I believe it will be more accurate or fraud-proof, but because any detected attempts to defraud this process with LLMs and agentic code will make the cheating blatant and undeniable. #NoAI