Rob MickeyDec 17, 2022
Jacob T. Levy

In case people don't know about this yet: AI text detector.

https://huggingface.co/openai-detector/

GPT-2 Output Detector

Dec 17, 2022 at 10:22pmWeb
Show threadÁlvaro La Parra-PérezDec 17, 2022
@jacobtlevy Someone expressed worries about false positives: https://wandering.shop/@janellecshane/109464780848248093
Janelle Shane (@[email protected])

Attached: 1 image Important note for educators about now: it looks like GPT-2 output detectors can detect #chatgpt output pretty well. https://huggingface.co/openai-detector

The Wandering Shop
Show threadJim OleskeDec 17, 2022

@jacobtlevy

FWIW, I just put in 4 AI answers that circulated on a law prof listserve back in November. This said 3 were "fake" (over 99% chance) and 1 was real (over 99% chance).

Interestingly, the one it said was "real" was the one that the professor who posted them to the listserve found to be the best of the 4 AI answers (the subject matter of the answers is outside my area of expertise).

Show thread(((Sarah)))Dec 18, 2022

@jacobtlevy

The GPT-3 chatbot flunked incredibly hard, at least on my 500-word essay on Gandalf and Saruman. I'm impressed; I really thought it should pass.