This July, OpenAI launched Study Mode, a feature that helps users learn through guided questions rather than direct solutions. A week later, Gemini launched ‘Guided Learning’ to help people learn through ‘probing and open-ended questions’.
Two major LLMs launched a learning feature just a week apart: are online courses the new thing AI’s replacing? Even the thought made us at Class Central go
To investigate this, I spent 10+ hours testing both platforms across four different subjects, from trigonometry to personal budgeting. I evaluated their abilities and their effectiveness, and here’s what I found: while AI won’t replace teachers yet, they can be great assistants.
Gemini Guided Learning vs ChatGPT Study Mode
I spent 10+ hours on both platforms, learning:
- Trigonometry: a school subject I never grasped
- Standard deviation: a topic in research that I wanted a refresher on
- Personal budgeting: a life skill I wanted to practice daily
- Generative engine optimization (GEO): a new professional skill I wanted to apply
I compared them based on three questions:
- Could they explain concepts clearly?
- Am I confident about my understanding of the topic post learning?
- Could I apply the knowledge in real-time?
Gemini’s Guided Learning
Gemini taught me trigonometry and standard deviation (SD) from core concepts (it refreshed my knowledge of right-angled triangles before introducing the basics of trigonometry). It broke down complex concepts into digestible parts.
This focus on simplifying broad topics helped me retain information, but it also made learning slow-paced. For skills like budgeting, I wanted quick tips that I could apply, but it took time to get to practical applications.
Gemini’s main weakness lies in its hallucinations. It asked me to look at images (to help me understand angles and sides of a triangle) while learning, but it didn’t produce any images. Plus, it didn’t inform me that my answers (for math questions) were wrong. In fact, it offered positive affirmations such as “That’s great!” for them. It also solved further questions based on my wrong answers.
Sometimes, Gemini would explain a concept but not follow up with questions. These questions offered me a direction, and without them, I felt lost.
But Gemini lived up to its “no answering for the user” policy. I asked it to write an essay for me, and it suggested we work on it together. Even when I pushed, it didn’t give in, which was impressive.
ChatGPT’s Study Mode
ChatGPT was like that efficient, no-nonsense teacher who might seem tough, but actually wants you to learn. Instead of vertical learning (going deeper into one topic), it was horizontal (teaching different topics related to the main concept).
The questions were directional — helping me decide what to learn further or assess how I want to approach my goals.
It was practical and intent-oriented, so it was more personalized. It was also more proactive. It made spreadsheets and even a playbook to structure my blog content for GEO. It also seemed intuitive and conversational, and I’d recommend it for practising light skills (skills that don’t require training, in this case, budgeting).
Sometimes, it leaped between subtopics. For example, while teaching trigonometry, it jumped from triangles to unit circles without explaining or warning about the change in topics. But when I asked it to slow down, it did.
Study Mode’s biggest flaw is that it does things for you. I asked it to write an essay, and it did, without probing or suggesting we work on it together (like Gemini did).
The Verdict on Study Mode and Guided Learning
Both tools have their strengths. ChatGPT excels at application-based learning, and Gemini at foundational learning.
But do they provide the comprehensive structure and confidence that expert-designed courses deliver?
The Big Question: Can LLMs Replace Online Courses?
While testing ChatGPT for therapy, Dr. Harvey Lieberman, a psychologist, said:
“I concluded that ChatGPT wasn’t a therapist, although it sometimes was therapeutic.”
It’s the same for learning. LLMs aren’t teachers or instructors; they are learning assistants, supplementary tutors, and answer providers.
The differences between both learning platforms and online courses are obvious — a lack of visual cues, structure, and direction. While these are crucial, there are deeper differences that need to be understood.
The Intent and Goal Fulfillment
When MOOCs began, millions joined to learn from scholars, grow with a community, and get Ivy League certifications. Dhawal, our founder, in his podcast with Coursera, spoke about this:
“You were getting access to a university course; the same course that somebody gets offline. You could do the assignments, and there was grading, homework, and a huge community. It was the allure of “Hey, you can take a course online from Stanford” or “You can take a course online from MIT”, and all these great universities.”
While the enthusiasm isn’t the same today, online courses stay relevant because the intent is the same — to teach.
LLMs started and marketed themselves as the answer to everything. But cases of cheating during exams and assignments revealed how easy it was to misuse it. In fact, just two months after ChatGPT had launched, a survey found that 90% of college students used it for homework. Even teachers and education experts considered it a problem.
OpenAI’s Study Mode seems like an answer to this problem, with its Socratic questioning (probing questions to develop critical thinking skills). While the intent of helping people learn with an assistant rather than an answer provider sounds good in theory, it’s also easy to switch to the normal mode and get your answer (side note: as I mentioned, Study Mode resorts to doing things for you, such as writing essays).
Easy access to instant answers means we use our brains less. This lack of mental exercise has weakened our thinking skills and lowered IQ scores in developed countries. Study Mode and Guided Learning claim to strengthen thinking; this means their (major) intent is to address problems that their own platforms helped create.
Personalization and Post-Learning Confidence
Both OpenAI and Gemini claim they offer personalized learning, and you could argue it’s something that online courses don’t offer. But personalized learning means approaching topics too narrowly, focused only on your immediate goals, rather than a comprehensive understanding.
When experts (usually experienced professionals) design courses, they aim to take you from introduction to understanding. During my evaluation, I tried to learn generative engine optimization on ChatGPT and Gemini for 2 hours. While I understood the concept (on Gemini) and how to practice it (from ChatGPT), I still wasn’t confident about my knowledge. I kept thinking: have I missed anything? Should I learn more about this?
I wanted a pre-determined structure, from introduction to application. So I signed up for a three-hour course on Udemy that explained its foundations, applications, and best practices in detail. I didn’t have to worry whether I had learned everything; I trusted the instructor.
Hallucinations, Trust, and the Human Effect
Online courses might not be tailored for your needs, but they have a personal touch that stems from an instructor’s/expert’s experience or years of knowledge. Having an instructor also adds accountability on the instructor’s part to deliver accurate and usable information.
When you see the tiny, almost invisible note that says “ChatGPT can be wrong”, it undermines its ability to offer reliable guidance. Plus, with LLMs hallucinating (which is a result of training them to sound confident and fluent when they can’t predict the next words, instead of admitting they don’t know the answer, according to OpenAI), we’re a long way from trusting their judgment.
When I’m learning, I like to know where I’m wrong so that I pay closer attention to the solution and retain it. On LLMs, I’ve gotten a “That’s great!” for the wrong answers too (and Gemini used my wrong calculation to solve the problem). This overenthusiasm is again a result of training.
Simon Willison, a programmer and the co-creator of the Django Web framework, tried to get the system prompt for Study Mode. See the “Tone & Approach” section? The platform is supposed to be “warm” and “patient”.
These words of affirmation, especially in the wrong places, get tiresome. Not every question needs a “That’s a great question!” or a “Perfect”. Teaching requires assertiveness and confidence that experience brings.
In Benjamin Breen’s analysis of OpenAI’s Study Mode, he says:
“They (LLMs) are like improv comedians, always ready to “yes, and…” anything you say, following the user into the strangest places simply because they are instructed to be agreeable.
But that isn’t what helps people learn. Some of the best teachers I’ve had were actually fairly disagreeable. That’s not to say they were unkind. But they had standards — a kind of intellectual taste that led them to make clearly expressed judgment calls about what was a good question and what wasn’t, what was a good research path and what wasn’t.”
So What Roles do LLMs Play in Learning?
OpenAI and Gemini are quick learning tools, but not comprehensive ones. I’d go for them when I want to understand a concept or practice a light skill (giving a speech, structuring an essay, etc).
But when it comes to mastering a skill or building one from scratch? They aren’t reliable enough or trained enough yet.
Until major issues such as hallucinations and a lack of impactful visual cues are resolved, I’d choose an in-depth course and learn from an experienced teacher over ChatGPT or Gemini.
The post ChatGPT Study Mode vs. Gemini Guided Learning, and Why Neither Replaces Online Courses appeared first on The Report by Class Central.
