Show HN: Sanctum – An LLM-Guided Learning App

1 point

JTan2231

a year ago

https://apps.apple.com/us/app/athenaeum-sanctum/id6744429496

It's in an infant stage as I figure out exactly how I want this.

In a more marketing speech style,

---

I love learning things and soaking in information, but I constantly run into 3 ergonomic problems with doing so:

- Quality study materials are often so dense as to require an extended session of focus

- Ergonomic study materials require too much work to find + aren't always flexible to what I want to learn/study

- Social media feeds have rotted my UI tastes to prefer short-form, transient interfaces that build familiarity with repeated, intermittent exposure

So, I'm making a flashcard app to address these points. In short,

- Reference-based deck generation

  - LLMs are great, but human sources are always best.

  - Give it a PDF or web page, and it'll reference that resource in creating your deck

 - AI-powered grading

  - The vector math powering LLMs allows for much more loosely structured answer grading than just naive keyword matching.

  - Know the concept, but can't phrase it in the colloquial terms? Don't worry, phrase it as best you can and it can tell how well you get the idea.

 - Natural deck progression

  - If you've been performing well enough on a deck, take a quiz to see if you're ready to expand the flashcards in your study.

---

There are a few things I still am looking to iron out:

- I don't think flashcards are the optimal UI for learning

  - They're great for memorizing + I do think improving pattern-matching is essentially learning is, but I don't think flashcards are the most optimal vehicle for meaningful learning

  - i.e., Life is more than just trivia

- Personalization

  - Right now, you take a quiz to expand the deck along the same lines as the existing subject matter. I think this is okay at best and could very much be improved.

  - One of the ideas I had initially around this is that LLMs can be used to traverse the embedding space of accumulated knowledge in a way that subverts/transcends the traditional academic taxonomization of studies. 

  - i.e., I think UI is the bottleneck in soaking in more information/ideas. I think LLMs do well enough in pattern-matching requests to information that they could act as librarians for the internet--then, it's just a matter of scrunching down the information into more convenient interfaces (e.g., mobile).

- Sourcing, references, etc.--I prefer something human-made at the other end of the tunnel, but I'm not sure how long that will last.

- The name--far too esoteric

This is an active WIP--would love to hear what you all think.

- LLMs are great, but human sources are always best. - Give it a PDF or web page, and it'll reference that resource in creating your deck - AI-powered grading - The vector math powering LLMs allows for much more loosely structured answer grading than just naive keyword matching. - Know the concept, but can't phrase it in the colloquial terms? Don't worry, phrase it as best you can and it can tell how well you get the idea. - Natural deck progression - If you've been performing well enough on a deck, take a quiz to see if you're ready to expand the flashcards in your study.

- They're great for memorizing + I do think improving pattern-matching is essentially learning is, but I don't think flashcards are the most optimal vehicle for meaningful learning - i.e., Life is more than just trivia

- Right now, you take a quiz to expand the deck along the same lines as the existing subject matter. I think this is okay at best and could very much be improved. - One of the ideas I had initially around this is that LLMs can be used to traverse the embedding space of accumulated knowledge in a way that subverts/transcends the traditional academic taxonomization of studies. - i.e., I think UI is the bottleneck in soaking in more information/ideas. I think LLMs do well enough in pattern-matching requests to information that they could act as librarians for the internet--then, it's just a matter of scrunching down the information into more convenient interfaces (e.g., mobile).