An Iowa school district is using ChatGPT to decide which books to ban. Official: “It is simply not feasible to read every book” for depictions of sex.::Official: “It is simply not feasible to read every book” for depictions of sex.
An Iowa school district is using ChatGPT to decide which books to ban. Official: “It is simply not feasible to read every book” for depictions of sex.::Official: “It is simply not feasible to read every book” for depictions of sex.
Yes exactly and that’s only one of the problems with this. Honestly, how does this even work? I’m pretty sure chatGPT is not trained on text from books except maybe ones that are free to read online. Even if it did, its’ training set ends in 2021. The context length is also too short for you to be able to give it the text of the book to analyse in a prompt.
Just…what? How does this even work? Maybe I’m missing something, I’m definitely no expert but both the policy and the implementation seem pretty broken to me.
The only way this could possibly work would be to send the entire text of every book into ChatGPT. If you ask ChatGPT something like “does the following passage contain depictions of sex: [text goes here]” it might work. However, ChatGPT’s context size is smaller than your average book, so you’d need to break it up into multiple requests.
Even with ChatGPT this is probably still not feasible to run against an entire library of books, so I would guess they’re half-assing it.
Technical feasibility aside, this is a stupid, evil idea. I hope everyone involved in this scheme gets fired.
I just asked GPT-4 if the Twilight Series is “suitable for children” and the output was pretty spot on - summarised five issues that might be problematic and then finished with “It’s always a good idea for parents or guardians to consult with a librarian or teacher”.
I bet none of this is based on the actual contents of the book - it would be based on independent reviews of the book. For example it highlights issues that “some parents” have raised. Which means you really are just getting a summary of what humans think of it.
Try it with an obscure book that doesn’t have any reviews online. Or with one that was released after 2021.
You’ll either get the response that it has no idea, or it’ll come up with a plot based on the title and evaluate that one.