By Alicia Gomez | UConn Journalism
Alexa Udell, a third-year psychology student, was having trouble in her statistics class. She was used to using tools like flashcards and notes to study for her psychology exams, but she found herself stumped on how to study for statistics. Trying to figure out the practice problems in her textbook felt like an impossible task.
Desperate, Udell turned to another solution: ChatGPT, an artificial intelligence natural language model. Finally, a weight lifted off her shoulders. She could study without feeling lost or panicked if the lecture went too fast for her to understand.
“I look at my textbook for practice problems and ask it to do that for me. It’s really nice because it does it step-by-step,” Udell said. “Like for a mathematical proof it’ll say stuff like: ‘Because of this, we do this. According to this rule, we do this. Here is the formula.’ ”
ChatGPT, an Artificial Intelligence chatbot trained to produce human-like text based on input, can spit out essays in seconds based on a prompt or code. It can even give advice. Although there has been a lot of concern about students using the new technology to cheat, there is also a growing trend of students and university writing centers using ChatGPT as a valuable tool in education settings.
Tom Deans, director of the UConn Writing Center, said that about 70% of UConn students surveyed were interested in learning more about using AI writing tools in their coursework, according to data he presented at the Northeast Writing Centers Association conference.
“I think everybody’s gonna be using it, It’s gonna be inside Word, it’s gonna be everywhere,” Deans said.
For example, Deans said ChatGPT has gained popularity at the UConn Writing Center, where tutors are experimenting with using it to help students with their writing assignments.
Deans said in his classroom that he allows students to use ChatGPT as a tool with credit. However, he believes professors should be clear in their course policies about what they define as plagiarism regarding ChatGPT. He said that if students use ChatGPT to write their full essay, it is already covered by UConn’s academic dishonesty policy. However, professors at UConn vary in their policies on using ChatGPT, Deans said.
“In my own classroom, I said, ‘You can use it for whatever you want, except for when I say you can’t,’ ” Deans said. “But you have to write an acknowledgment statement that says what and how you used it.”
Deans said he has seen professors prevent students from using ChatGPT primarily by changing their assignments. For example, they may write questions that rely on specific information students may learn in class or account for more recent readings, as ChatGPT is only trained on information up to 2021.
Even though ChatGPT has been an invaluable tool, Deans said it could not replace face-to-face tutoring offered in writing centers. Human tutors offer specific advantages ChatGPT won’t be able to replicate.
“It doesn’t understand context all that well, so it’s not going to understand your particular class and where you are in your writing process and what you are trying to do,” Deans said. “So that can only happen in a conversation.”
ChatGPT also tends to make up false information that sounds correct when it is unsure of the answer, said Alexander Solod, the President of the AI club at UConn, who has been working with Deans. Solod said this phenomenon is called “AI hallucination.” Because of AI hallucinations, students may have to be careful about using ChatGPT on their assignments. For example, if a student plugs in their essay question and ChatGPT spits out a response, it may sound like a plausible answer to the untrained eye, but a professor may easily be able to see that it is riddled with factual errors.
However, despite the disadvantages, there is still usefulness to be gained with ChatGPT, like its availability, Solod said.
“This right here is like a 24/7 tutor that is skilled in almost every single at least undergraduate university task,” Solod said.
Solod and Noah Praver, a UConn Writing Center tutor, found that ChatGPT can expand paragraphs, give tips, write a thesis, think of a title, rewrite sentences, and more. However, it is important to make specific prompts, or “prompt engineer.” They believe that professors and tutors should train students to do that.
For example, when Solod struggled with understanding the lectures and textbook for his biochemistry class, he told ChatGPT to take on the role of a biochemistry tutor to explain what he was stuck on, he said. Solod was also able to ask ChatGPT to come up with questions that a biochemistry professor would ask on a quiz based on Solod’s notes.
“By assigning it a role and making your question really, really specific, you could get the most out of it,” Solod said. “By assigning it a certain ‘personality,’ you are able to extract more information or to extract a better result out of the model based on whatever task you want to accomplish.”
Solod predicts its popularity will continue growing as Google and Microsoft integrate generative AI technology into its writing tools.
“These tools will become a staple in our day-to-day lives kinda the same way that Google has become ubiquitous and everybody uses it,” Solod said.
Solod also recognizes risks associated with trying to prevent the use of ChatGPT in the school setting. He has been seeing a growing number of Reddit posts where students claim that they have been falsely accused of using AI tools in their writing because AI detection tools are largely inaccurate, he said. So he has tested it out himself, putting in both middle school assignments he wrote long before the release of ChatGPT as well as the Declaration of Independence. ZeroGPT detected both of these texts as AI-generated.
“The want to want to try to detect and regulate is one that we will slowly need to try to bypass and instead focus on how to teach responsible and ethical use of AI instead of a blanket ban on it,” Solod said.
A UConn student in his freshman year studying computer science who has asked to go by Rafi for fear of academic repercussions said he knows ways to evade detection from plagiarism detectors and professors when he uses ChatGPT. One way is to be specific in prompts, similar to prompt engineering.
“I can ask the ChatGPT to write it as if it’s coming from a college freshman, and ChatGPT will write it in that style,” Rafi said.
He said that students could also take the output they receive from ChatGPT and reword it.
“You put it in ZeroGPT, you see if it’s detected, you could change it up a bit and just keep going through until it’s much less or completely undetectable,” Rafi said. “If you put in spelling mistakes, then that really throws ZeroGPT off.”
However, Rafi said he suspects that detection tools at ZeroGPT are not popular among professors at UConn yet, but may be in the future.
Rafi said he believes there are two reasons why people may use ChatGPT instead of doing their own assignments. First, some students may be just “going through the motions in college” and are not making the most of their education; they are simply just trying to pass. So it may make it easier for these students. Some students are under academic pressure and want the time to focus on other classes, Rafi said.
“As advice for students, look, if you get away with it, if it’s a class you don’t feel is necessarily going to help you, do it while you can, take advantage of it, sure,” Rafi said. “Just don’t be stupid. Give it good prompts, read over it, understand what it’s saying, try to at least understand a little bit. Don’t just be lazy, that’s all.”
You must be logged in to post a comment.