Tested papers returned AI scores as high as 69%.
Interesting --- someone had their NeurIPS Position Paper desk-rejected after being flagged by Pangram, claims they ran recent papers authored by the Position Paper chairs through Pangram, and got scores of 69%, 45%, 36%, and 24% AI-written.
Distribution shift is an important concern, especially as second language speakers may actually learn how to write from AI.
Interesting --- someone had their NeurIPS Position Paper desk-rejected after being flagged by Pangram, claims they ran recent papers authored by the Position Paper chairs through Pangram, and got scores of 69%, 45%, 36%, and 24% AI-written.
Original post: https://old.reddit.com/r/MachineLearning/comments/1tvwctd/neurips_used_uncalibrated_ai_detector_for_desk/
Interesting --- someone had their NeurIPS Position Paper desk-rejected after being flagged by Pangram, claims they ran recent papers authored by the Position Paper chairs through Pangram, and got scores of 69%, 45%, 36%, and 24% AI-written.
Tested papers returned AI scores as high as 69%.
Interesting --- someone had their NeurIPS Position Paper desk-rejected after being flagged by Pangram, claims they ran recent papers authored by the Position Paper chairs through Pangram, and got scores of 69%, 45%, 36%, and 24% AI-written.
Users in the replies expressed respect for linking to old Reddit in coverage of NeurIPS desk-rejecting position papers with an uncalibrated AI detector.