I’m taking a high-level course in a specific scientific field. For my last exam, I found a way to put all the course materials into one document. I transcribed the lectures, copied text from PowerPoint slides, pulled info from public Quizlets from people who’d taken the course before, and used feedback from previous exams to figure out how best to study.
I then used LLama2 on my own computer to organize and categorize all the concepts that came up the most. Afterward, I asked the program to make a practice exam focusing on the most likely content.
I studied it for a couple of hours and ended up scoring 139 out of 140… Was that machine learning? I feel like I almost cheated. Did I get lucky? Does anyone have suggestions on how to make this more efficient or know of any programs that already do something similar?
What challenges should I expect with this approach?
You’re using tools that are based on Machine Learning (which is a big category), but the key is how you compiled your materials. Gathering rich, high-quality data and asking the right questions helps make studying easier. You did a great job by focusing on likely exam content.
Some challenges:
There’s a risk of missing steps or not gathering the right info.
AI’s summarization tends to leave out important details.
For example, if your lecture materials weren’t detailed enough or if your Quizlets didn’t cover enough ground, the AI might miss some key points. You’d still need to rely on textbooks for those cases.
And, when the exam isn’t just about factual recall—like if it includes essays or problem-solving—AI-generated materials alone might not be enough. You’d still need to study those areas manually.
But this is a solid approach! Keep experimenting with different prompts, or analyze past exams to improve how the AI helps you study.
@shamaris
Thanks for your response! I agree there are limitations with this approach. The most helpful thing was probably having access to past exams—it really helped predict the structure of the questions.
If I give the AI too much unnecessary information, I think it could mess things up.
The test was for a physiology class covering acid-base and electrolyte balance in the renal, urinary, and respiratory systems. It had about 50% multiple choice, 20% true-or-false statements, and the remaining 30% was short essays and a diagram to draw.
Do you think I should use a different model for this type of exam?
That’s really impressive! AI worked as a tool to gather and organize data, which allowed you to study faster and more efficiently. The key here is that you used AI to help, not replace, your efforts. Many tools try to replace the human part, which can cause issues like ‘AI hallucinations.’ You’ve used AI to complement your studying instead of trying to make it do everything.
A big limitation is that the model can’t improve unless you retake the same class. AI can only work with the data you give it, and every course is different. So while you can reuse the AI, every class will still be unique based on how the professor teaches.
I’m not aware of any tools doing this exact thing, but AI tools are still early in their development. More apps will be available soon.
Would you mind if I shared your approach as an example?