There will be a few different types of Students using chatGPT. Those blatantly cheating and just turning in blindly these works.
Those who have a hard time putting their ideas to page, but when asked can talk through their ideas.
A third group may arise, those who cheat blatantly, but then parrot what they used.
What am I saying here though, oral exams will separate these groups. Identifying what group the pupil is in helps understand how to actually help them progress.
My bias for this idea comes from my mother is a retired college professor and gave oral exams for two decades with large success in weeding out those who couldn't formulate ideas of their own.
What’s the point of learning proper English when chatgpt can do your homework…
ChatGPT is amazing but getting someone else to do your work is not novel.
I’m also quite certain schools will implement an AI-detection mechanism, and wouldn’t be surprised if OpenAI were to expose that as a product offering.
As for the viability of an ai detector, if a human can’t recognize the difference in a fundamentally human dataset (language) there is no reason to suspect ML would be better suited to it. A watermarking scheme would be more effective.
We used to trade bags of weed for student metro cards. Others trade lunch tickets for homework. I saw a student hand over their north face jacket and a cell phone to have someone else to take an AP exam for them. They later told their parents they were robbed on the way home.
As Jeff gold bloom would say: “life uhhh… finds a way.”
Re: watermarking, it’s exactly what I alluded to by “OpenAI May even expose this as a product.”
Make the pupils read, study, etc. then they get x hours in class to write their paper with only pen and paper allowed.
You can also do both and then compare. If what's produced at home is very different from what's produced in class...
(Or if you can assemble a formula sheet and pull a b after copying answers from the back of the textbook then you earned it)
