HN Reader

I'm surprised (but maybe shouldn't be) that one of the students in the article had to prove her humanity for a written assignment in a comp sci class. You'd hope a comp sci professor of all people would know how unreliable these "detection" tools are...

An acquaintance of mine is in college for CS and he says his classmates use AI for everything and don't really learn the material. He dreads group projects because he'll be paired with classmates who have AI write code that doesn't work and they don't understand. But also it's apparently lead to a lot of writing code by hand on paper for assignments and tests and stuff and he is also annoyed that he'll get dinged for stuff like overlooking a semicolon that would be an obvious compile-time error when writing code on paper.

I feel bad for him because it seems like it's going to be hard for him to find a job next spring. (And in all fairness, I don't know how good or bad he actually is at CS.)

So soon there will be programs to generate papers keystroke by keystroke to mimic human drafting. It’s an arms race

How do you prove a negative beyond ethical codes of conducts with onerous penalties for violation? Continuous screen and webcam recording during work?

This is just the story that keeps on getting written and won't go away.

Yes, students are going to cheat, they always have. If you are a college professor and you don't want to reward cheaters, make large percentages of a student's grade come from in person tests with no electronics allowed. Simple.

Tell the students up front at the beginning of the course. They're almost surely 18 year olds at least and legal adults.

What a college shouldn't do is outsource "detection" (much less in-source it), which is impossible to accomplish at an adequately low False Positive rate.

A professor at my day job offers oral exams to students to avoid this - zero takers.

Good, I always hated writing papers and the whole arbitrary requirement of x pages was always so stupid to me

Really? You don’t think someone could get their point across in 8 pages just as well as 10?

Hope this means you just have to write more in person and are graded on the quality of your answers vs quality + quantity

I know a professor who has everyone except 1 student completely failing one of her classes because they all used AI to write their papers. Most of them didn’t check the output and have things like “like and subscribe” in the middle.

The real headbanger was how many people wouldn’t write proper sentences and would skip capitals, grammatical marks, and spelling on the in-person exams.

These people graduated from high school with a high enough grade to go to college.

To say I didn’t believe said professor is an understatement… but I know them and they have no reason to lie. They don’t know what to do because they are getting pressure from administration to pass the students.

The balance between teaching the use of modern tools, but not letting that ruin learning seems to be real, based on a Master's thesis that was handed in yesterday. It can help you get a more academic wording, but language rewriting is probably the most likely to get caught in one of those cheat detectors.

Hopefully the research is solid enough that it's easy to dismiss any automatons complaining about LLM language...

I have kids in middle and high school, and this is all getting figured out in real time. I think we're going to end up with a system that requires personal trust, physical presence, paper tests, handwritten work. Maybe it's for the best; I've been disappointed with most "technology in the classroom" efforts. It all seems worse than paper and pencil which was fine for hundreds of years. Nowadays though handwriting is harder for everyone; me included.

I feel like the solution to this problem is just blue books and in-class assignments.

Higher education that is not a part of a PhD program is effectively obsolete. You don't need to pay to learn. In the past, technical textbooks could be difficult to digest and understand, but now LLMs can dumb it down to one's level. Demonstrating your open source projects and contributions based on their star count, essentially a proxy for their impact, is better.

I was helping my young teen with her English homework recently and I kept thinking that I might given her more feedback a few years ago. "Maybe you could clean up this sentence...". Now I think it's good that the writing looks unpolished. It's going to stand out among all the elegant perfectly worded essays with LLM tell-words like "delved" and "seamlessly".

I don't know where this ends up.

TAed last semester and I caught about 8 students for cheating, all time high out of 6 semesters (never really caught anyone before). All of it is pretty unsophisticated since I don't think we even could police sophisticated methods that well, you know the legendary printing the exam notes on the coca cola label and other high effort methods. Most of them were actually whispering to eachother in exam or otherwise blatantly staring at answers from another student, in a room with 4 proctors. Another case for an essay sort of assignment we saw straight up copy and paste of entire paragraphs from another student in their class, and for another case from past materials. I don't know how they thought they could get away with any of this. Both of these students emailed actually surprised they failed the assignment....

Really our detectors don't check AI use at all. You can probably chatgbt it all and I couldn't really tell. However, they check if this sentence matches with a sentence either out there on the internet in the primary sources or, most often, from another students submitted assignment from our or a partner university using the same saas. And people are getting caught and having their college career ruined just from that, straight up copying and pasting with no paraphrasing even and thinking we won't notice that.

They must have gotten away with highway robbery in highschool during the COVID years. Employers will have to look at this COVID generation with a big side eye and put more emphasis on in person tests of knowledge. Maybe they should use the various testing centers used for MCAT/GRE/LSAT/etc for these remote interviews where they provide a locked down desktop and have cameras surveilling the room.