I played with this a while after uploading one pdf chapter from the textbook I teach out of. The chapter covers text segment and stack dissection of simple C programs, as in rudimentary reverse engineering.
I was happy with many of the flash cards, but I struggle with the idea of encouraging my students to try it because I got 2 or 3 bad cards in the span of about 5 minutes. One tried to ask about an analogy in the text, but it reverse the order, butchering its meaning. One stated hat GDB would be useful for exploring logic and syntax, but not runtime, errors. And another posed a distinction without a difference, which would result in a lot of confused students and, consequently, busier office hours.
As a student I've been researching the use of AI for various tasks and have come to the initial conclusion that AI improves quantity but not quality such that a student is almost always better off spending an extra 30-60 minutes producing materials themselves.
A lot of tasks like note taking are essential to knowledge retention such that even supplementing notes with AI generated summaries creates clear tradeoffs that personally I choose to avoid.
I'm more curious in AI use by teachers to produce better media for student consumption, paired with tools like Figma, Adobe, Microsoft office, and other tools. In my experience teachers are forced to produce a lot of content, akin to content creators, without the proper incentives, quality controls, or training. I think AI can help there.
I think when paired with great tools, like Figma, then the quality/quantity trade-off of AI is changed. Perhaps there is a lesson there that can eventually be realized by students, but I haven't seen it yet.
In my undergrad engineering department, we see the biggest opportunity for improvement as being in lab sessions, where it's not reliably apparent who is wrong or struggling, and it takes a while to explain to each group what is wrong. We've had some luck feeding from student answers from our lab instructions into structured AI prompts that provide individualized feedback.
But that takes even more time to work on student material, which is not a highly rewarded way to spend one's time. Edit: plus, most of the time my experience is that lower-friction assignments result not in better-spent study time, but in less study time for my course.
Notetaking in handwriting seems to be especially powerful for retention. Even when I had the option to type my notes in college, I kept on hand writing them (despite wishing since like middle school that I could type notes since extended writing with a pen cramps my hand).
Fun fact: The founder of TikTok Alex Zhu worked on an education app but failed so miserably he decided to pivot to musical.ly, which later became TikTok.
I assume this is a generic LLM + RAG, so you can't prevent hallucinations and there's a chance (which gets better as the subject gets more complex) that the study materials will be wrong.
And anyway, creating flash cards is a huge study booster by itself. It doesn't make sense to automate that step away.
It makes a ton of sense. I've been doing my own custom version of this for the past year and it's a HUGE shift in studying efficiency.
- LLMs are getting better. That's a reality LLM haters need to contend with. The versions that didn't solve your one issue perfectly 2 years ago simply aren't in the rotation any longer. A major release with significant improvements is a nearly monthly occurrence.
- I don't know how to convince anyone of this, but the hallucinations have literally zero impact. Not only are they rare and getting rarer, but because you can always spot a hallucination you learn to ignore them.
- As someone who's a huge fan of flashcards, you're wrong that this is worse than making your own cards. Instead of spending hours and hours making cards, those hours are spent actively engaging with and absorbing more material.
I can't speak to OP's product, but with my system it's the closest thing I'm aware of to downloading information directly into the brain.
I like this concept, but I am concerned about copyrights. Public information might not be as big of a deal as a textbook or closed PDFs. What happens to the data after?
I am encouraging people to insert their own custom notes. but even if they add textbooks or anything I think it will be fine as they are not distributing the data and just learning for their own in a fun way
Hey it's working great! Small nitpicks while I'm testing on my macbook + chrome.
- It can be easy to "miss" the next card because scrolling feels really fast
- I accidentally uploaded the same files twice, but can't figure out how to delete the duplicates.
it's just a demo. you can create a subject and then insert your study materials then you can quizzes will be generated which you can scroll like tiktok but in this you are actually learning
I was happy with many of the flash cards, but I struggle with the idea of encouraging my students to try it because I got 2 or 3 bad cards in the span of about 5 minutes. One tried to ask about an analogy in the text, but it reverse the order, butchering its meaning. One stated hat GDB would be useful for exploring logic and syntax, but not runtime, errors. And another posed a distinction without a difference, which would result in a lot of confused students and, consequently, busier office hours.