#teachers and #school persons: I can appreciate that #ChatGPT is a significant leap. But I can also appreciate that this GPT2 detector easily detects GPT3.5 outputs. Things are going to change, but some perspective is warranted #education: https://huggingface.co/openai-detector/
@DavidKnuffke That is cool, but I would not rely 100% on it; indeed, I tried on a few outputs, and while some are detected as <1% real, others are considered >90% real. I haven't, however, found a piece of text written by a student that was classified as AI-generated (however, I have only tried a few).
I would be extremely careful about relying on such a tool as "proof" of academic misconduct.
Most of all, I think that discussing these tools with students is extremely important.
@nicolaromano definitely agree with all you say here. I haven’t put in a text piece with >> 50 tokens that hasn’t come back either clearly fake or real. But I’m sure they exist. For me it’s more about proof of concept. This kind of thing can be automated and appended to other programs (ex email clients, word processors) to signal possible origin.
Also agree that talking about these things with students is crucial.