A whole new ballgame in cheating. Introducing ChatGPT

[Sun_Worshiper]

https://www.chronicle.com/article/gpt-4-can-already-pass-freshman-year-at-harvard

That says more about Harvard than about ChatGPT.

Not really. Students at any humanities or social science program in any university are assigned tons of essay assignments that Chat GPT is perfectly capable of passing. Harvard is not unique and it is obviously better than the vast majority of schools.

https://www.chronicle.com/article/gpt-4-can-already-pass-freshman-year-at-harvard

That says more about Harvard than about ChatGPT.

Or anyway, says more about the sorts of assessments that colleges have been using.

Until very recently there was nothing wrong with giving writing assignments to students. I imagine most of us will continue to do so, even despite the ability of ChatGPT to pass them with ease.

I'll be changing my approach in the fall to have more presentations and more in-class writing with the lockdown browser. But I'll also have takehome essays, which I'll make nuanced enough that students will need to do some work on their own, even if they are using ChatGPT. I'll also teach students the strengths (which are impressive) and weaknesses of text generative AI.

[Hegemony]

I read an amusing tidbit today, which is that in French ChatGPT is pronounced like "Cat, I farted" ("chat, j'ai peté"). Which is causing the French no end of hilarity.

[apl68]

I attended a session on AI at a recent library association conference.  An academic librarian spoke of having gotten requests for books and articles that didn't actually exist.  The patrons had used ChatGPT to conjure up a citation list, and it had hallucinated convincing-looking citations for imaginary sources.  It's awkward when a patron asks you to locate an item that doesn't actually exist!  That said, the speaker also said that most of the time the bot's citations are legitimate.

[apl68]

Just yesterday I was doing a web search for information about the history of book covers and found what I think was a product of a ChatGPT hallucination.  It informed the reader that the earliest books were bound in human skin, and that this remained common until the 17th century.  Books bound in human skin do exist, but they were never a remotely common phenomenon. 

Actually that search turned up quite a few weird-looking collections of images and words that all had a lot of suspicious overlap with each other.  It looks like online searches are getting increasingly contaminated by goofy bot-generated content.  Yet more noise-to-signal ratio, as if we didn't have enough already.  We discussed at that library conference meeting mentioned above concerns that the situation will further deteriorate as the bots feed on bot-generated content to fill the internet with even more bot-generated content.  A couple of us used the term "ouroboros." 

[Hibush]

It looks like online searches are getting increasingly contaminated by goofy bot-generated content.  Yet more noise-to-signal ratio, as if we didn't have enough already.  We discussed at that library conference meeting mentioned above concerns that the situation will further deteriorate as the bots feed on bot-generated content to fill the internet with even more bot-generated content.  A couple of us used the term "ouroboros." 

The "training set" or "corpus" used by generative AI is so critical to the result that is produced. The nature of that set appears to be highly protected by the companies, so we users and corpus producers don't know. High quality materials are copyright protected. Using them in a training set for a commercial product is not likely to be fair use. Lawsuits coming!

It is easy to imagine cut-rate genAI shops using a corpus of free stuff that is fairly arbitrarily chosen. They would then generate more free corpus to build and sustain an iterative cycle of deepening BS. Whether that cycle is an ouroboros or requires a different metaphor is for literary scholars to establish.