Personal development and better infrastructure for learning, "Anki V2"

Science & technology

Brent Burdick

ActiveGrant

$10,000raised

$10,000funding goal

Fully funded and not currently accepting donations.

Project summary

I’m a 21 year old college drop out and self-taught programmer. I’ve spent the past year working with startups and making side projects, mainly on the frontiers of LLM powered software. I’ve recently become interested in learning tech and spaced repetition systems.

I’ve forgotten most of what I learned in high school, and I regret not having a way to commit things to my long term memory. I think spaced repetition is extremely powerful for learning, and it has been overlooked and not popularized as much as other technologies in the past 10 to 20 years because it doesn’t fit well with the quickest methods we have of distributing new technologies (venture backed businesses).

I think the recent advancements in LLMs have made learning tech ripe for disruption. I’m looking to significantly improve some part of the learning experience with spaced repetition / Anki.

A few ideas on my mind:

Making unstructured texts into working Anki cards to study.
Improving language learning by having people practice cards in contexts as opposed to recounting specific words
1. Intention to generalize this past language eventually.
Creating a chat interface which could:
1. Ask you about a particular card or topic in natural language
2. Update the scheduling of cards based on your inputs
3. Answer questions about cards you don’t understand
4. Create new cards for you based on your questions
Improvements on an SRS algorithm which allows for accurate updates before a particular card is due. For example, in language learning, you’ll review an article hundreds of times in the first week.
Allow for “fuzzy” answers. There are many possible correct answers to a particular problem. For instance, when grading a translation to a sentence, it would be great to accurately grade every possible correct translation as correct, rather than just one, hardcoded translation.

What are this project's goals and how they be achieved?

The chief aim is to make a meaningful and lasting contribution to learning tech / the SRS community.

A meaningful contribution is one that significantly improves the experience / results of SRS for some set of users and would not have been made by normal enthusiasts and weekend hackers otherwise.

For instance, I don’t think simply combining ChatGPT and Anki in a primitive or unthoughtful way would be that meaningful, because people have already done similar things, and it’s not that different than just switching back and forth between Anki and ChatGPT. Instead, if I created a method for parsing questions and turning them into new accurate cards, or was able to feed the questions into the SRS algorithm and improve it that way, then this would be the sort of contribution I would consider meaningful.

Of all the ideas I have to move forward, I’d like to do a deeper analysis of what’s achievable and impactful, and what’s been tried before, and then quickly build and test my improvements in the real world.

As a second, stretch goal, I’d like to generate revenue with the advancements that I make, so I can afford to keep working on this project in the long run.

How will this funding be used?

This funding will be used for about 6 months of runway to spend time exploring and building this project as opposed to a full time software job.

Who is on the team and what's their track record on similar projects?

I am a college dropout, and have worked at a variety of venture funded startups in the past year with LLM powered applications: including https://www.knn3.xyz/, https://www.atlaszk.com/, https://morph.so.

I’ve also built and launched personal projects: https://twitter.com/BingBongBrent/status/1679565807809069066 and https://twitter.com/willdepue/status/1670099641218764804.

I know what it’s like to take a project from nothing to something, and have done it multiple times. I’ve failed plenty as well.

What are the most likely causes and outcomes if this project fails? (premortem)

The most likely cause of failure, I think, is taking a big bet on a particular direction and that direction ends up being wrong and I spend too much time building in that direction without testing / measuring its usefulness.

What other funding is this person or project getting?

None.

Austin Chen

about 1 year ago

This grant falls outside of our more established pathways, but I'm excited to approve it anyways, as a small bet on a people-first funding approach (where I think the regranting mechanism shines).

I'm a bit baseline skeptical of SRS/anki, having seen tools-for-thought people push for it but fairly unsuccessfully -- eg I was very excited for Quantum Country but it doesn't seem to have gotten wider adoption, nor personally helped me very much. However, I would be excited to be wrong here, and it's possible that LLMs change the game enough for there to be a good angle of attack!

Brent Burdick

about 1 year ago

@Austin Hey Austin! I definitely hold some of the same reservations as you here, and have been similarly skeptical. Yet, I have to admit that a few of them have actually been very impactful on me / the world. I'm curious what you think makes the difference between the projects that succeed and those that don't or if you have more to add there.

Austin Chen

about 1 year ago

@Brent It's not clear to me what successful examples are... which have been impactful for you? I think foreign language and MCATs are two domains where SRS have proven its worth, but outside of that those memorization-heavy domains, the flashcard approach hasn't become popular. It's also damning that most successful people don't rely on SRS, afaict.

I think there's something definitely interesting about the core observation of SRS - "learning happens via repeated exposures to the subject, and we can program that to our benefit." But it also seems to me that "flashcards" are a dead end, UX-wise, given all the research that has gone into them for relatively little adoption. I think there's a lot of space for innovating on other interaction models -- eg in what ways are social feeds like Twitter a spaced repetition system? Or Gmail?

One other random note - for a while, I've wanted a SRS/anki thing that helps me stay on top of my various personal contacts (friends, acquaintances, etc). "Making friends" is a domain which lines up neatly with exponential backoff, I think -- it's easiest to make a friend by spending a lot of time with them in the beginning, and then staying in touch gradually less and less over time.

Brent Burdick

about 1 year ago

@Austin The impactful ones have really been Anki (and memorization heavy stuff like it, to your point) and Notion I might count as a "tool-for-thought" as well, although sort of in a different domain.

I also agree that "making friends" could perhaps benefit from SRS-style thinking. I started an Anki deck for this purpose actually hahaha, but the UX is hard to nail, and I haven't ended up using it much.

I think you make a lot of interesting points. Thanks for sharing.

donated $10,000

Isaak Freeman

about 1 year ago

Main points in favor of this grant

Spaced Repetition is one of the best tools out there to accelerate the learning pace and retention of anything. Anki, the most commonly used app, works fairly well. But it's mostly old software, and GPT4 opened up huge, underutilized opportunities for accelerating learning pace. I wonder why we still don't have automated flashcard creation, personalized feeds of content from a textbook, personalized explanations of concepts, etc. I'd be excited to see these integrated into a couple of great tools.

Brent has been working on this and generating lots of ideas. He thinks that "Anki v2" is overdue. He's dropped out of university to self-teach software engineering, and is currently exploring what to do next. In the spirit of investing in people, not projects, Brent is a quick learner, very thoughtful, and quick to get stuff done. I'm excited to support Brent in whatever he's doing next, but especially his focus on improving learning tech has been promising.

Donor's main reservations

It's possible Brent ends up failing to build something that's notably better than Anki or doesn't make good use of his next months of exploration.

Process for deciding amount

Common career redevelopment grant size intended for 3-6 months of exploration, new projects, etc. (Cf. Emergent Ventures, ZFellows)

Conflicts of interest

Important to note, Brent is a housemate and friend. Though, this allowed me to see him make rapid progress on his initial projects and made me confident in recommending this project.