
Understanding Value of Learning Fuels ChatGPT’s Study Mode
Instead of generating answers, Study Mode asks students questions about what they know.
Photo illustration by Justin Morrison/Inside Higher Ed | SDI Productions/E+/Getty Images
When classes resume this fall, college students will have access to yet another generative artificial intelligence tool marketed as a learning enhancement.
But instead of generating immediate answers, OpenAI’s new Study Mode for ChatGPT acts more like a tutor, firing off questions, hints, self-reflection prompts and quizzes that are tailored to the user and informed by their past chat history. While traditional large language models have raised academic integrity concerns, Study Mode is intended to provide a more active learning experience. It mimics the type of Socratic dialogue students may expect to encounter in a lecture hall and challenges them to draw on information they already know to form their own nuanced analyses of complex questions.
For example, when Inside Higher Ed asked the traditional version of ChatGPT which factors caused the United States Civil War, it immediately responded that the war had “several major causes, most of which revolved around slavery, states’ rights, and economic differences,” and gave more details about each before producing a five-paragraph essay on the topic. Asking Study Mode the same question, however, prompted it to give a brief overview before asking this question: “Would you say the war was fought because of slavery, or about something else like states’ rights or economics? There’s been debate over this, so I’d love to hear your view first. Then I’ll show you how historians analyze it today.”
Study Mode is similar to the Learning Mode that Anthropic launched for its chat bot Claude for Education back in April and the Guided Learning version of Gemini that Google unveiled Wednesday. OpenAI officials say they hope Study Mode will “support deeper learning” among college students.
While teaching and learning experts don’t believe such tools can replace the value faculty relationships and expertise offer students, Study Mode’s release highlights generative AI’s evolving possibilities—and limitations—as a teaching and learning aid. For students who choose to use it instead of asking a traditional LLM for answers, Study Mode offers an on-demand alternative to a human tutor, unbound by scheduling conflicts, payment or feedback delays.
But in an economy where generative AI’s ability to gather and regurgitate information is threatening the future of entry-level office jobs, students will need to understand what they’re trying to get out of their college coursework if they want Study Mode to deliver on its promise to foster deeper learning.
“Over all, it does represent a significant departure from how people are normally encouraged to use generative AI, given that the value proposition of this technology is largely couched in the concept of efficiency, speed and offloading things that are not as important,” said Trey Conatser, assistant provost for teaching and learning at the University of Kentucky and director of UK’s Center for the Enhancement of Teaching and Learning. “But Study Mode actually asks us to stick with it for it to work, take some time to sit with the friction of learning and be patient with the practice and feedback.”
Critical Thinking in Age of AI
The release of Study Mode comes nearly three years after the debut of ChatGPT—and the many other large language models that have followed—set off ongoing concerns among educators that college students’ increased reliance on traditional LLMs to write papers or solve math problems may erode their critical thinking skills.
There’s not yet a consensus on if generative AI helps or hurts critical thinking, but at least two studies published in the last year found that adults, including college students, who use LLMs to research and learn about a subject exhibited weaker reasoning and analytical skills. And those may be the exact types of skills the roughly two-thirds who are already using generative AI tools like ChatGPT will need to make Study Mode reach its full potential.
While ChatGPT’s Study Mode is programmed to query students rather than generate answers, the users themselves also need to be prepared to push the LLM to ask deeper, more nuanced questions. At least, that was Conatser’s experience when he tested out Study Mode over the weekend and asked it to help him study for a hypothetical quiz on Romantic poets from 1780 to 1820.
First, the LLM asked him more about his education level and the focus of the quiz. He told it that he’s a college junior whose quiz will focus on historical context and specified that he wanted help studying William Wordsworth’s poem “The Prelude.”
The following conversation, which Conatser shared with Inside Higher Ed, ensued:
Study Mode: Wordsworth described “The Prelude” as “a poem on the growth of my own mind.” What exact subtitle did he eventually attach to the poem that captures this idea?
(Give it a try, and we’ll check your answer before moving on.)
Conatser: That’s not so much what I will need to answer. It will be open book, but I will be asked to provide some interpretations to historical context and literary traditions.
After some more dialogue asking Study Mode to focus on historical context, it generated the following question:
Study Mode: “Bliss was it in that dawn to be alive …” Based on that word, “bliss,” what general emotion do you think he felt when he first encountered the early French Revolution—excitement, fear or apathy?
Conatser: That seems like an obvious question
Study Mode: Understood—let’s turn up the difficulty. In “The Prelude” (Book X), Wordsworth writes of “how in my heart a dreary weight / Fell, and I turned away …”—a moment that marks his break with the revolution he once celebrated.
Question (one step): Which single, headline-making event in France during 1793 does this passage point to, and why did that event feel like a betrayal of the revolution’s original ideals for many sympathetic British observers?
Although it took some prodding to get the LLM to ask the type of questions Conatser wanted to see, “I’m still in the driver’s seat with Study Mode,” he said. “That’s where students are going to need to be when they’re using it.”
Knowing when to push back against Study Mode when it’s not asking the types of questions that will help students master material will be key to helping them get the most out of its ability to foster critical thinking, Conatser said.
“If someone is going to be totally passive with Study Mode, it may not drive them to where they need to go and they’ll still need to be thinking about how they’re supposed to be studying, what kind of assessment they’re leading up to and what kinds of skills will be more important for that assessment,” he said. “If you don’t know what you don’t know, then you won’t be very effective at using these tools even if they have the types of guardrails that Study Mode has.”
And if a student isn’t motivated to use Study Mode to its full potential, they can prompt it to generate complete answers, just like a traditional LLM, according to Christian Carmody, a senior researcher at the Research Institute for Learning and Development, who also tested it out this week.
“For a current events assignment, I prompted Study Mode right away and told it, ‘Before we engage, I do not want your help with this or [to] encourage me to think through this. I do this on my own another time. I really just want the answers,’” Carmody recalled. “It did exactly that.”
The ability for students to easily manipulate Study Mode could add more pressure to colleges and universities that are facing growing skepticism from students about the value of degrees in the age of AI.
“Students should be able to think about why learning is valuable to them and why they should be able to engage with material in a way that’s challenging and force deep thinking,” he said. “Until a student has that mindset, I’m not confident that they are going to use this study and learning tool in the way it’s intended to be used.”
Source link