One thing is the 58 000 that have to re-take the exam. I’m just as interested in the 2% of exam-takers that were terminated due to “unspecified conduct issues”, and how many of those were unfairly thrown out for cheating without having done anything wrong…
I think you’re confused about the AI involvement here.
AI was used to check for cheating in a remote exam. The percentages that passed were several times more than in previous years, while 2% got tossed for “unspecified conduct issues”.
Assuming those 2% were actually cheating and not the AI making a mistake is the issue.
These were multiple choice exams. There was no ability to copy/paste from an LLM. Could they have used another computer/monitor to look things up? Absolutely, but hardly relevant to the issue.
There are also people who grew up in Ghana /Nigeria who are being accused of submitting chatbot writing because the fine tuning was done by their compatriots.
I mean AI is totally fine for fine-tuning. Helpful, even. It’s a fabulous proof-reader. The issue is using it for a first or second pass, or having it make suggestions for changes.
I believe the point was they were treated as having used ai because ai sounds like them because it was trained by people with their accent.
Similar to the “This is not ai, I’m just autistic”.
One thing is the 58 000 that have to re-take the exam. I’m just as interested in the 2% of exam-takers that were terminated due to “unspecified conduct issues”, and how many of those were unfairly thrown out for cheating without having done anything wrong…
sure, keep selling this victim trope online. Idiots are submitting papers with LLM prompts still in them.
You realise that it is a statistical impossibility that there were absolutely no mistakes?
I think you’re confused about the AI involvement here.
AI was used to check for cheating in a remote exam. The percentages that passed were several times more than in previous years, while 2% got tossed for “unspecified conduct issues”.
Assuming those 2% were actually cheating and not the AI making a mistake is the issue.
These were multiple choice exams. There was no ability to copy/paste from an LLM. Could they have used another computer/monitor to look things up? Absolutely, but hardly relevant to the issue.
There are also people who grew up in Ghana /Nigeria who are being accused of submitting chatbot writing because the fine tuning was done by their compatriots.
I mean AI is totally fine for fine-tuning. Helpful, even. It’s a fabulous proof-reader. The issue is using it for a first or second pass, or having it make suggestions for changes.
I believe the point was they were treated as having used ai because ai sounds like them because it was trained by people with their accent.
Similar to the “This is not ai, I’m just autistic”.
How is that relevant to the AI falsely identifying students as cheaters?
Sounds like they aren’t the ones haven’t done anything wrong. Unless you know that’s everyone that commenter is talking about