I’m sorry, but that’s not true at all. It’s pretty easy to check too. Even if we ask it some bog-standard English class essay, I dunno, “How does Huckleberry Finn develop morally throughout the book,” ChatGPT isn’t going to produce all obviously plagiarized answer. Not in the way that a plagiarism detector can detect.
2 Likes
Myself and a team did a quick research project early last year where we mixed up ChatGPT answers to questions with a random sample of student solutions to the same questions and had them blind graded by experienced markers.
As part of the process we put all the texts through standard anti-plagiarism software like CopyCatch and TurnItIn and none of the synthetic text was flagged as being copied from elsewhere. In fact, it tended to show greater levels of originality than student text. We didn’t use any of the then extant AI text checkers because of their dubious user privacy and copyright policies*.
The human markers said the AI text was often more rambling and less focused than the human text, but it was broadly correct and in most cases it would score a modest passing grade. It completely fell apart with postgraduate level text and deep knowledge - such as looking into a research paper (not possible at the time because of the sandboxed nature of ChatGPT then) or doing a reflective activity on previous learning.
- in a later test, we put 10,000 student solutions through TurnItIn AI and found a persistent few percent of solutions flagged as AI despite having been extracted from archived presentations before LLMs were available to the public - which suggests these tools aren’t all they’re cracked up to be. Given these tools can’t explain their decisions, we decided to stop using them for conduct cases. Nine months later, I’m still waiting for the journal to get back with a final yes or no on that one - not that it’s a fast moving fields or anything.
18 Likes
If by “this” you mean generative AI that can write entire, and passable, essays, I see what you mean. But then, I’m still a bit cloudy on how those who have trouble fulfilling an essay-writing assignment benefit by having ChatGPT write one for them?
5 Likes
It’s not hard to spot work that a teacher might strongly suspect was written by AI. But suspecting and having hard evidence are two different things, especially if the teacher has to make a decision about whether to give the student a failing grade on the assignment.
8 Likes
This seems like the sort of trick that would work most of the time; but has some too-common-to-dismiss failure modes that would leave you with outcomes that are tricky to deal with.
The one most likely to have actual legal or policy teeth would be screenreaders: tiny white text on a white background is significantly less visible than black on white, if you can see; but they are both equally visible you are are sufficiently visually impaired to be relying on a screenreader to handle the text for you. Doesn’t mean that the blind kid isn’t also potentially cheating; does mean that they’d ‘see’ your ‘hidden’ directions as clearly as the rest of the assignment and might well act on them even if they weren’t.
Anyone who copy-pastes without preserving source formatting(as is a plausible move if you want your onenote or whatever to remain coherent when constantly pulling things in from multiple sources) to aggregate all their assignments from their respective instruction pages into some sort of planning or calendaring application is also pretty likely to unveil the text. Again, nothing particularly implies that an organization-tool user might not also be an adherent of technlogical workflow optimization generally; but it won’t be a good look to hold up your smoking gun when they can show you that the ‘hidden’ prompt comes right over when you create calendar events to keep track of your assignments.
This would get more into the weeds; but I suspect that at least some of the stylesheet overwriting that slightly nerdier browser users have been using to make the web sane again for ages would also reveal the concealed, as those tend to clobber a site’s more…idiosyncratic…design choices deliberately, in order to make them usable for people who aren’t a degenerate cabal of branding types.
More mainstream options like the “reader mode” in safari might well do the same thing, depending on exactly how the formatting is handled, what CMS you are going through, etc.
5 Likes
This was (most of) your post that I was responding to, in which you were responding to @ClutchLinkey’s statement:
So I didn’t mean ChatGPT does all the work, but rather, is one step in a larger process. Research and editing skills are still very much needed, for example.
4 Likes
I think it is different, however: someone who works from a transcript rather than a live lecture is much more closely analogous to someone who dictates to a speech-to-text system rather than typing. I’m very much on board with the requirements of alternate input or output methods, after years and years of having assignments dragged down just for atrocious handwriting; but it seems like there’s a fairly profound difference in kind between interfacing with the information differently on either input or output and doing completely different work.
2 Likes
How to spot AI generated content:
Spotting AI-generated essays can be challenging, but certain characteristics can help discern their origin. Firstly, inconsistency in writing style and tone throughout the piece may indicate AI involvement, as human writers typically maintain a consistent voice. Additionally, AI-generated essays may lack depth or coherence in arguments, often presenting surface-level insights without meaningful analysis or original thought. Unusual syntax or grammar errors, especially when combined with overly complex vocabulary or unnatural phrasing, can also be indicative of AI composition. Furthermore, a lack of personal perspective or cultural references may suggest machine-generated content, as AI lacks personal experience or cultural understanding. Finally, rapid production of high-volume content on diverse topics, especially if the quality remains consistently average or better, may signal AI involvement due to its ability to generate content at scale.
ChatGPT wrote this.
(You’d know it was me if it had grammatical errors that I caught an hour later. Supid brain.)
5 Likes
Look at that Viktor Frankenstein. Just look at him.
5 Likes
Blücher!
5 Likes
ChatGPT and other AI sources, like Wikipedia before them, are adequate places to start a creative effort, but should not be mistaken for the finished product, at least in this iteration.
5 Likes
Would you stop it? The horses are skittish enough!
4 Likes
People who didn’t read the book think that Frankenstein is the monster’s name, people who read the book know that Frankenstein is the doctor’s name, people who understood the book know that Frankenstein is the monster.
14 Likes
You are Dr. Cleo Markham’s mother, and I claim my £5!
4 Likes
Yeah, no shit, it was obvious from the style halfway through the first sentence. Which is an interesting data point in itself.
9 Likes
Right? I caught on right away too.
6 Likes
I love this, and it feels like the right approach.
It reminds me how during my engineering education cheap programmable calculators became a thing, so the good profs would tell us “the assignment questions are set up so that you could use a simple calculator or a full software package on a computer, but either way you have to understand what you’re trying to do.”
I also love the banana / Frankenstein approach as a very quick “did you even read it?” check.
5 Likes
just have them hand-write all essays in class like an essay test? don’t give them the topic ahead of time but let them take as long as they need and have proctors working in shifts if need be. they can bring in whatever books they want (or they could be provided,) if they don’t know the questions ahead of time they can’t write crib sheets in them. take everyone’s phone at the door and circulate to check for smuggled phones. they can write outlines and drafts on looseleaf and then use best handwriting in the bluebook.
5 Likes
I like it, but it would definitely be something one might accidentally fall prey to. I use OneNote for things and assignments, and I keep all of my notes and assignments in it, and remove the formatting when I paste things in (so I can add notes, like answers to review questions or to-do items on assignment requirements), which would make the text appear. I’d probably assume “oh, this is clearly some plagiarism prevention thing, so I should definitely include it”.
Mind you, I suspect a few questions by the instructor about it would clear things up, but it doesn’t strike me as being completely without potential false-positives. It’s inventive, though, and I do admire that.
7 Likes
description for SEO: Explore the cutting-edge world of AI with this stimulating video from AI Basics, your premier destination for all things artificial intelligence. In this fascinating discussion, a brilliant educator shares her unique strategy for detecting when students submit papers written by AI – using specifically planted “Trojan horses” in her prompts. The video gives a glimpse of a simple but innovative idea: embedding a distinct phrase that must be included in the essay, which proves if students have used AI to complete their work or not. It’s a captivating peek into the ever-evolving world of technology in the realm of education. But the intriguing insights don’t stop there! We reveal how this approach goes beyond just students and assignments. Dive into the discussion to discover how technology, such as AI like ChatGPT, is employed by recruiters in assessing job applicants. Experience the mind-bending future of AI technology with this video from AI Basics. It’ll leave you questioning the role of AI in every facet of our lives, from classroom to the corporate world. Join us in delving deep into the real-world applications and implications of AI. Tune in to AI Basics channel now, where we bring AI down to earth for everyone to understand!
… this alt text is longer than the video 
2 Likes
