The AI Text Classifier is a fine-tuned GPT model that predicts how likely it is that a piece of text was generated by AI from a variety of sources, such as ChatGPT. This classifier is available as a tool to spark discussions on AI literacy.
Very glad to see this on your radar! In addition to subjective classifications like "likely" "very unlikely" etc, it would be useful to have the ability to get additional statistics that might share the actual probabilities (ie for 3rd party apps) and even more useful, highlight parts of the text that have the highest probabilities.
Amazing work! Looking forward to seeing the impact of AI Text Classifier in the AI literacy discussions!
Report
Definitely going to be useful for teachers if it can stay ahead of the curve but with just a little human tweaking it does start to break down a bit. I assume the longer the text that was generated the more likely it is to recognize a lot of it as AI.
Also not everything has to be written by a human, so I think scoring how much AI wrote text is a very contextual tool. Like, Gmail has been prompting for years... how many people use "In addition..." now as an opener? Is that good or bad? What about when emails start getting scored inlined? Will that make you trust a communication from a known human more or less if they use AI in an email that probably didn't need to be written by them anyway?
Regardless ChatGPT has teachers freaked out so I think it's nice to have tooling like this they should be able to access for free.
Replies
Charma
SocialBu