Texas professor failed half of class after ChatGPT claimed it wrote their papers | Hacker News Reader

Texas professor failed half of class after ChatGPT claimed it wrote their papers

businessinsider.com

21 pointsCharlesW22 comments

> However, ChatGPT was not designed to identify content created by artificial intelligence — including its own. There are several other programs, including Winston AI and Content at Scale, that are able to do this.

That claim to be able to do this.

This arms race has barely gotten started.

Here is the update from the student as of yesterday:

"Situation is (mostly) resolved

In a meeting with the Prof, and several administrative officials we learned several key points.

It was initially thought the entire class’s diplomas were on hold but it was actually a little over half of the class

The diplomas are in “hold” status until an “investigation into each individual is completed”

The school stated they weren’t barring anyone from graduating/ leaving school because the diplomas are in hold and not yet formally denied.

I have spoken to several students so far and as of the writing of this comment, 1 student has been exonerated through the use of timestamps in google docs and while their diploma is not released yet it should be.

Admin staff also stated that at least 2 students came forward and admitted to using chat gpt during the semester. This no doubt greatly complicates the situation for those who did not.

In other news, the university is well aware of this reddit post, and I believe this is the reason the university has started actively trying to exonerate people. That said, thanks to all who offered feedback and great thanks to the media companies who reached out to them with questions, this no doubt, forced their hands.

Allegedly several people have sent the professor threatening emails, and I have to be the first to say, that is not cool. I greatly thank people for the support but that is not what this is about.

Also heard from professor that his job may or may not exist after today due to his foul language and unprofessional communications with students but not due to the AI accusations.

Finally, the prof issued an apology to the 1 student exonerated so far and it appears the school is well aware they are not yet equipping to deal with AI in an academic setting, and this will be a HUGE learning day for not just A&M commerce but the system as a whole. My goal for today is to ensure all the other students receive exoneration if they so deserve."

Exoneration implies a presumption of guilt, whereas the default should be a presumption of innocence, especially in a scenario where no concrete evidence can be furnished.

The professor has undoubtedly caused a great deal of stress and anxiety for a number of students who may or may not have cheated, and did so in an unprofessional way. They should, of course, be fired or censured in some manner for this.

This professor should be fired for stupidity. The bar for being a professor has to be higher than "i pasted your stuff into ChatGPT and it said it wrote it, hence you get zero".

It's worse: he didn't even know its name, thought it is called ChatGTP.

Imagine paying 60k a year (out of state) for this. Huge black mark on the reputation of Texas A&M.

Amusingly, the students put the professor's email into ChatGPT and it claimed to have written that as well.

Also various journal papers the professor wrote (or, rather, claims to have written!).

Just plugged this comment in, also claims to have written your comment!

Just have the professor do the same with a random sampling of his own writings. It’ll falsely claim that it wrote half of those too.

Also the professor should receive a failing grade for utterly failing to understand how any of this works.

This seems a bit pointless. After using ChatGPT for so long to do writing in professional context (documentation, description of how things work etc.), so that I can reduce my cognitive load, as these things are not my forte and I always struggle to write. I learned that on many occasions, I would write something exactly as if ChatGPT would have written it. Basically, brain learns and adapts, seeing patterns. In a few years, you probably won't know if something was written by AI or by human as everyone will be writing in the same style.

This doesn't make sense. People have been writing for a long time. If that was true we would already all write the same

Well actually we are all writing with the same words, grammar rules, proverbs and so on. It is only that some creative minds try new stuff all the time, adding variety again.

The people writing with the help of GPT will eventually start writing like GPT without GPT. Makes sense?

I guess much like he asked of his students, the professor should not have used ChatGPT to do his own work of determining which students cheated using ChatGPT :)

If I were a university student, how might I protect myself from being falsely accused of using ChatGPT?

The cynic in me says that human student papers will stand out for being badly written, while the ChatGPT cheaters will have more superficial polish (if less solid reasoning).

Type your work in a Google Doc with edit history enabled. Or if you're writing code, push signed commits to a repository shared with your professor, every few hours you're working on it. Honestly, even for essays, maybe share the Google Doc with your professor ahead of time. It might annoy them but at least you're covering your bases early.

Alternatively: write about controversial topics and/or fill your paper with obscenities :)

Do what the exonerated student did - use the track changes feature in your word processor. I would think something like Github and some simple tools (so a non programmer could use easily) to provide basic version control would be a good way to track versions and time stamps of updates to docs, projects, etc.

Ask for exams in examination rooms, no computers allowed.

Use revision control and frequently commit your work, preferably along with comments describing your thinking process. In law school I used Git as that's what I was familiar with and it made it easy to backup my work (git push), but I suppose the tracking feature in MS Word might suffice.

In law school you cite almost every sentence in a written work. In theory you don't need to cite your own, bare opinions, but even then it's common to provide citations for rhetorical effect. And legal citations are very specific and descriptive, which is why footnotes are often so long. Rigorous citation is just a good idea in any field, and they're more costly to fake, at least if they're more than just last name & year. (If you're playing a long game, make sure archive.org has a snapshot of any online sources; they provide browser extensions for that.)

It also might not be a bad idea to capture your browsing history, so you can show a timeline of how your research went. In law school this history is already captured by the legal portals as part of their billing system, but standard browsing history would be more than sufficient.

Finally, and perhaps most importantly, if you do any of this, then be consistent. Don't do this for one class but not another, or have browsing history for some papers but not others. Consistency speaks to both credibility and reliability of evidence. So it's better to have no browsing history for any than only for some. A professor in a Wills & Trust class explained it this way: one day you'll be presented with a document you helped execute 30 years ago and asked, for example, if the identities of witness signatures were verified. You'll have no direct recollection of this. But you could answer in the affirmative; you could say, "Yes, because whenever executing a document I always did X, Y, and Z, without fail." (Or alternatively, "I almost always did X, Y, and Z, and the rare exceptions are memorable".) And if you indeed were consistent about this, there'll be corroborating evidence. So if someone is skeptical about whether your Git version history was authentic, you want to be able to point them to version histories for all your most recent papers as corroborating your process.

Threaten to sue. Then, sue.

Reply on news.ycombinator.com