Page 1 of 1

An AI-driven Study Tool

Posted: Mon Aug 24, 2020 1:34 pm
by ezubaric
You may know us for building QANTA, the quizbowl AI system (and side projects for playing QB online, writing questions free of stock clues, etc.). We’re building on our previous systems to not just have computers play trivia games but to help humans play trivia games better. To do this, we’re introducing KAR3L, a flashcard system built to make studying quizbowl (or anything really) more efficient!

KAR3L uses spaced repetition, a concept that may be familiar through other flashcard software used to study for quizbowl, like Anki and Mnemosyne. For those unfamiliar, spaced repetition is a method where a fact is shown repeatedly over gradually increasing intervals, as a user grows more familiar with the information.

While existing applications work decently for learning and reviewing a fixed set of facts, they have no knowledge of flashcard content. Facts are scheduled on an individual basis and existing scheduling algorithms are slow to adjust for a difficult question or variations in user ability. Previous systems also generally select new facts at random. Instead, KAR3L incorporates knowledge of flashcard content to suggest new facts and schedule old ones. For example, if you answer a fact about George Washington correctly, KAR3L intelligently estimates your knowledge of the Revolutionary War era to suggest new, harder related facts to study. In addition, KAR3L uses machine learning to account for variations in pre-existing knowledge and memory capacity between people.

While we think this will both help people study and lead to better studying experiences, we need to prove it scientifically. For that, we invite you and any interested friends to use KAR3L, which can be accessed at KAR3L can also be installed as a desktop, Android, or iOS app! Instructions can be found at

We encourage users to spend at least 10 minutes a day using the program, but this is not a strict requirement and users can study as little (or as much!) as they would like any particular day. $20 gift cards will be given to the five users who spend the most time studying in the app over the course of the study, which will last until the end of June 2021. 10 other users who answer over 200 questions will be randomly selected to receive $10 gift cards.

If you have any questions, suggestions, etc. please contact Jordan Boyd-Graber at [email protected] or Matthew Shu at [email protected].edu or comment below! You can also join our Discord!

Re: An AI-driven Study Tool

Posted: Sat Sep 05, 2020 12:25 pm
by Nonameentered
Following some recent updates, KAR³L now loads flashcards to review, statistics, and leaderboards significantly faster! For those who previously encountered some issues with these features or the sign-up process, I encourage you to give the app another try. We also welcome feedback and suggestions in this forum thread, particularly observations about the scheduler.

In addition, we have open-sourced the application part of KAR³L, which can be found here: For those with Github accounts, bug reports are encouraged to be reported in the Issues tab. In addition, we welcome contributions to the project and provide detailed instructions for setting up a local copy of KAR³L for development. For those who have trouble setting up the development environment, feel free to contact me for additional help. We also expect to open-source the scheduler portion of KAR³L sometime soon but there is not ETA yet for when this will happen.

Re: An AI-driven Study Tool

Posted: Sat Nov 21, 2020 3:46 pm
by Nonameentered
Hey everyone! We're reaching the end of our first phase of experimentation (the app will continue to be available for use) and we'd like you to fill out this survey regarding your experiences using (or not using) the app here. Ten users who fill out the survey and study over 200 cards during our upcoming second phase of experimentation will be randomly selected to receive a $10 Amazon gift card. Regardless of how frequently you've used the app, we'd like to hear your feedback! If there's a barrier preventing you from using or trying out KAR³L despite having an interest in doing so, please fill out this survey too.

We've also published a blog post with a progress update on our research findings here: We hope you’ll take the time to read our post and talk about it in our Discord, this forum post, or by contacting us directly at [email protected].