Orbit: an experimental platform for small tasks repeated over time | Hacker News Reader

Orbit: an experimental platform for small tasks repeated over time

github.com

388 pointsgoerz57 comments

Hi, all. Author of the project here. Sorry to have caused some confusion: this is part of a research project, and it's not yet meant to be terribly legible to a public audience.

In the near-term, this project is about augmenting written texts so that readers can easily remember the key details. See https://quantum.country for an example or https://numinous.productions/ttft for detailed background and discussion.

Hi Andy, thank you for doing this. I've been checking your Twitter feed daily to see when this is released. I love the direction of your research and am really excited to see how Orbit develops.

I've been experimenting with integrating Orbit (before this release) into some Intro to Data Science Jupyter notebooks I use to teach a class (code, text, Orbit questions). I'm hoping to publish the course on the web later this year, and, if I can get a good integration, experimenting to see if adding SRS elements helps student performance.

I haven't looked at the code base yet, but I'm curious to see if I can integrate the questions with our code execution environment, and have spaced review of executable code inline in the question/answer prompt card.

Some context: The author created https://quantum.country which is a text book augmented with spaced-repetition. It seems that Orbit is a toolkit for creating these types of flashcard/spaced-repetition learning webpages for anyone (with the user data being stored at Orbit's server).

I'll have to say that I'm still not 100% sure how useful this approach is. Most of the times when you're learning something you're not merely reading, but you're also actively using the knowledge in some way. If I'm learning about B-trees then the major part of that learning process is implementing them. If I'm learning some complicated math I will solve a bunch of exercises. Where does flashcards in a text book fit into this?

It was also my impression (from the Anki community) that spaced-repetition is far more effective when you create the cards yourself. A common "beginner mistake" is to download a pre-made deck and start memorizing from scratch.

I think the "beginner mistake" is that you can download a pre-made deck of 1000 German words, many of which you have never heard before, so it will be very hard to make them stick in your memory.

With Orbit, I think the idea is that the prompts are very well crafted (Andy has some research on making good prompts) and also that you are encountering the questions right in the context where you are learning about them, so you have something to attach them to.

I think this is a complementary thing for implementing B-trees or solving exercises. That is needed as well, but will you still remember the finer details of splitting nodes in a B-Tree a month later? Or six months later? Which might or might not be necessary.

> I think this is a complementary thing for implementing B-trees or solving exercises. That is needed as well, but will you still remember the finer details of splitting nodes in a B-Tree a month later? Or six months later? Which might or might not be necessary.

This is a very good point. This is the type of knowledge which I don't care about retaining. I'm pretty confident that I can re-learn it in a few days when I need it later. It's also not critical to have this knowledge at the top of my head as I'm rarely going to be in a situation where I need to recall this immediately.

This is quite different from vocabulary where it is critical that you can find (or understand) the word in seconds otherwise you'll fail at communicating.

Fully agreed! I think the existing options for pre-made decks often resemble your "1000 German words" example, and the space of providing good pre-made decks is explored far too little and dismissed too quickly in the Anki community.

I can see something like Orbit very likely existing in the set of my future learning tools. Most texts that I read (doesn't matter with how much concentration), flow right through me and nothing sticks. The only way to make them stick is by exercising with spaced repetition flashcards, but creating them often takes so much time that I only bother to do it for the most important subjects.

A possible future evolution of the concept where cards unlocked on two independent websites can unlock further cards like "how does concept A (from website A) relate to concept B (from website B)" would be amazing, and seems very much in the realm of possiblity.

You make a great point that using knowledge is important in actually learning the study material. Personally, those moments of learning happen when I stumble across a problem where I'm applying my knowledge in the wild.

What I've found is that spaced-repetition software lets me keep in touch with information long enough to be able to have those moments more consistently, otherwise I read something interesting, take notes, and two weeks later its pretty much out of memory. By extending the amount of time info is living in my head, I get more chances to use the knowledge and take it to heart!

This doesn't lend itself to everything perfectly well though, I've had luck learning how to do household tasks like gardening, cooking where I'm semi-regularly getting chances to use what I'm learning

"Where does flashcards in a text book fit into this?"

Answering a question doesn't have to rely only on rote memorization. For example, a book might explain how to do addition, and then ask "What is 110+257?". You might need to use a piece of paper to work out the answer. And, if that same question is asked again a couple of weeks later, it's unlikely you'll remember the answer. You'll need to work it out again, reinforcing your understanding of the technique.

"spaced-repetition is far more effective when you create the cards yourself"

Yes, but part of the reason is that there's little point in having a card for something you've not encountered in another context. But Quantum Country doesn't present questions to you until after you've encountered the concept in the text.

> It was also my impression (from the Anki community) that spaced-repetition is far more effective when you create the cards yourself. A common "beginner mistake" is to download a pre-made deck and start memorizing from scratch.

Varies between people and subject matter I think.

All of my Anki usage thus far has been for language learning, and for this use case pre-made decks have been better than self-made for the simple reason that these decks are higher quality than I have patience to build myself. I do still plan to build my own decks for vocab beyond core 2-3k but for getting off to a running start I think premade is just fine.

Same here, I learned Chinese largely by studying a pre made deck of sentences with audio over the course of a few years. I had lots of conversation practice too but the SRS studying bootstrapped my knowledge to the point where I could even attempt conversations. I used to believe that using a pre made deck is a waste of time just because I heard a few people say so, now I know it’s not always true.

I made a SRS deck for PHP stuff when I was going to take the certification.

What I found was that studying the deck didn't help me at all. What did help was making the deck.

IMO, if something makes the deck for you, it's going to drastically reduce the usefulness of the deck. Making the deck means a lot, and possibly much more than studying the deck could do for you.

It does help that this is making it from something you're trying to read... But I think removing that effort is actually harmful to the experience.

As I understand it, the Orbit platform presents the cards to you within the context of the text you are studying, where you'd probably want to create a card yourself if you were making them.

For example you just read a section of a text and decide you should make a card for it. In Orbit the text would provide the card pre-made.

I could also imagine extending this to make it possible for the reader to create personalize review prompts, on the fly, and build up a personal card deck for the text.

That was my assumption, too. But it's taken the work out of it. You no longer need to think about the card you're making. You probably just click a few times and poof, there's a card.

There's no memorizing it long enough to put it somewhere else. There's no typing (or hand-writing). You don't have to re-read it more than about once, unlike hand-making SRS cards where you'd read it at least a few times.

Given that Andy says Orbit is currently a research project, perhaps the efficacy of SRS with pre-made cards v personally-written cards is something that could be studied more closely.

Research shows that doing the recall is the important part. The effect is by now well accepted. Off course if you do the cards yourself you can guarantee both their quality and their relevance to you. Ive heard of research pointing that learning things that are relevant to you helps. (If you are learning Spanish to speak with a relative or friend a flash card with an obscure literary word is probably more annoying than helpful). early episodes of Learning Scientists podcast talk about using vs creating flash cards if you want to know more.

Since it doesn't look like this project has been discussed on HN before (though it was a bit hard to search for: https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...), I've changed the title from "Orbit is now open source" to one that describes the project.

Edit: ah, but the author's related essay has been discussed a couple times:

Develop Transformative Tools for Thought - https://news.ycombinator.com/item?id=26078115 - Feb 2021 (1 comment)

How can we develop transformative tools for thought? - https://news.ycombinator.com/item?id=22147212 - Jan 2020 (44 comments)

How can we develop transformative tools for thought? - https://news.ycombinator.com/item?id=21152542 - Oct 2019 (74 comments)

Just recently I was reading about supermemo, a tool that’s about 30 years old but has been updated and focuses on improving retained memory. It was developed by Piotr Wozniak who based on this 2008 article has taken the system as far as humanly possible by surrending his life to the algorithm.

https://www.wired.com/2008/04/ff-wozniak/

The supermemo algorithm has been published and you’ll find all sorts of implementations including the Ruby gem forgetful.

I read the mnemonic memory article and I didn’t see what differs between orbit and supermemo, but that doesn’t mean the approaches are the same.

(Orbit author here) Wozniak is the pioneer, and like all in this space, we're heavily inspired by his work. Orbit is a vehicle for exploring some different ideas in this space. For instance, we're looking at how spaced repetition might be woven into a written form, like a book, so that authors can facilitate readers internalizing ideas deeply with relatively little extra effort. (see https://quantum.country for an example)

Are you familiar with Niklas Luhmann’s “zettelkasten” methods? Seems like there might be some fertile ground at the intersection of these two ideas...

Indeed. For example, org-mode can easily implement both a Zettelkasten and SRS. There are ready-made packages for both.

Andy (the Orbit author) has his own very interesting Zettelkasten-like set of public notes. In this page, for example, he compares/contrasts his "Evergreen note" approach to a Zettles in a Zettlekasten: https://notes.andymatuschak.org/z4AX7pHAu5uUfmrq4K4zig9x8jmm...

Additionally, his system does have spaced repetition built-in, too:

"I’ve embedded a Spaced repetition memory system into my notes: The mnemonic medium can be extended to one’s personal notes. This has substantially changed the medium, but in ways I don’t yet understand very well."

I do not believe he has open-sourced his note taking system, but it is fun to browse around in; he presents a lot of interesting ideas, particularly if you're into this space.

> I do not believe he has open-sourced his note taking system

He is using Bear, you can find on youtube how he creates those notes.

The schematics for the older versions of the SuperMemo algorithm (SM-2 and SM-5) have been published but they're quite old and any good reimplementation is actually a fork that fixes glaring deficiencies in the old algorithms.

The newest versions (SM-19) are proprietary and rely on a bunch of training data (they use some kind of ML) gathered from SuperMemo users (there is a way to get access to SuperMemo algorithm but you have to negotiate a license from memory). There is some rough outline of the algorithms on the SuperMemo wiki, but you couldn't reimplement it any more than you could reimplement Google Search given the Wikipedia description of PageRank.

> there is a way to get access to SuperMemo algorithm but you have to negotiate a license from memory

Well, SuperMemo may be of some help to the memory portion :-)

Is this public information? Where can I read more about SM-19?

Sorry I meant SM-18 (released in 2019, hence the confusion). But for completeness the only information available on SM-17 is one of two incredibly long articles written by Piotr Wozniak[1,2]. SM-18 is only a minor improvement over SM-17, with some changes to how difficulty is calculated[3].

[1]: https://supermemo.guru/wiki/Algorithm_SM-17 [2]: https://www.supermemo.com/en/articles/history [3]: https://supermemo.guru/wiki/Item_difficulty_in_Algorithm_SM-...

I read the readme, now can someone explain what this is like I'm stupid?

This page explains it a bit more, and gives a demo:

https://docs.withorbit.com/

It looks like a flash card widget that you can embed in your blog. Honestly I have no idea what the context is for needing this, perhaps the authors could point to a real-world use case? I don't understand what the "Forgotten" and "Remembered" buttons do, or why a new question pops up when I click one. It seems to loop forever through different cards in a random pattern. This page doesn't do a great job of explaining the project from a 50-foot level, and the homepage is even worse at that.

Are you familiar with Anki?

I found the docs as well. The docs jump right in and use words like "prompts", "prose", and talking about writing a bunch. I guess maybe this is intended for authors? The inspiration listed is a digital book (https://quantum.country/). Some context sure would be nice.

Agreed, none of the documentation does a good "elevator pitch" or summary of what the product is, or the intended audience.

neat font though.

The digital book is made by the same guy, Andy M. and Michael Nielsen. Orbit seems to be a platform to enable anyone to build similar content, with mnemonic content (flash cards basically) built in.

Yes, they really try hard to make it as obscure as possible. Here is my take. Imagine you publish an article on some complex subject matter - the kind that requires a pause and reflection after each sentence. Imagine that after a paragraph or two there is a widget where you can test your understanding of the above paragraph. It seems useful to me, but of course it depends on the actual implementation.

As whimsicalism says, it's like Anki.

You get the question and then the answer; if you remember the answer, the question won't be re-asked for a longer time, if you don't, the question will be re-asked sooner.

It appears (?) to be a sort of global tool---you embed your flashcards into the text you are writing and the reader gets asked your questions in addition to questions from the math book they read last week or last month.

It's a thing for learning stuff that fits on cards.

Perfect, thanks!

Haha, you're welcome!

decentralized, right in the readme. what more do you need to know?

I haven't used it/had no idea about it until reading this post, but here's my attempt.

You aren't stupid, it's just that the genre of projects (spaced-repetition systems) Orbit is in is pretty niche.

It mentions that it more or less wants to be a more generalized version of Anki. Anki is basically an application that goes, "Hey. Here are some flashcards you haven't seen in a while. Do you still know the answers?" The specific timing of these depends on how many times you've gotten them right in the past, with the aim of getting the information into your long-term memory.

Anki is incredibly useful for memorization. However, because of how limited it is, it's only really gained popularity in the medical student space and the language-learning community. It's not really extensible beyond what you'd expect from flashcards that you can embed video in.

What Orbit seems to claim to do is have a very extensible system for spaced-repetition, so that the underlying principles can be used for a wider array of things. However, it doesn't appear to be aiming for mass uptake; rather, it's mostly a research vessel to gain an understanding of the problem domain.

I'm really stupid and/or lack the attention for this, but thank you for trying!

> because of how limited it is...

What is so limiting about Anki? Anki cards are also fully javascript capable, and anything you can't achieve with just js you can do with their python addon api. From what I've seen the only use case Orbit serves that Anki doesn't is the ability to embed cards on your own pages.

I don’t think it’s the tool itself, rather the technique. I used SRS to learn kanji and it was nothing short of miraculous. I also used it to study extremely efficiently for well-defined examinations (e.g. proficiency exams for some basic college courses).

However, I’ve also tried SRS on other subjects with less success. One challenge is that in many knowledge domains, it’s not entirely clear what is worth remembering. Sure, you can purge cards that turn out to be unimportant, but the process of creating the cards in the first place can be awfully time-consuming so it starts to feel less efficient.

Orbit looks like a way to publish content with accompanying spaced repetition. Spaced repetition is a technique to efficiently improve long-term recall.

The most accessible primer on spaced repetition I've seen is Nicky Case's interactive exploration at https://ncase.me/remember/

I really wish people would stop dual-licensing with permissive licenses. This particular case seems pretty bad, and completely negates the whole point of releasing it under the AGPL in the first place, by having the BSL convert to Apache License in three years. Three years is no time at all, and after that it's like the AGPL never existed.

I don't agree with releasing libraries under a permissive license, but that's of course the prerogative of the author. Businesses don't need to have their GPL fears coddled, they can and need to learn how to work with copylefted software.

I have lost count of how many projects called Orbit I have seen so far.

This is the one that would have helped you remember the others.

Hadn't heard of SuperMemo / spaced learning before and has been awesome reading about it tonight, especially relating to personal note-taking. That said, one thing about QCVC I'm not wild about (so far) is the amount of meta content / sales pitching it leads with. Like, I get it, you wanna convince people of the value of this approach, but I'd prefer a version that just did the thing (teach the basics of QC) instead of spending so much time talking about teaching you the thing. Its like if I was reading a textbook and it spent half of the time talking about why I should do / engage with the exercises... ideally the method / process should be self-justifying and not need to make a case for itself. I get that QCVC was geared towards proving the concept, so I guess just hoping if this starts getting used by other authors, they implement just the process rather than spend an inordinate amount of time talking about it.

So this is the system described in [0], which I've been curious about for a while, but wasn't readily available. It'll be interesting to see how this compares or complements tools like Dendron. The style also reminds me of the fedwiki.

[0] https://notes.andymatuschak.org/About_these_notes

I saw the name and was hoping he was open sourcing his notes platform hehe.

This is interesting, but definitely above my head.

I'm excited by all of the neat hypertext developments we're seeing out there! Like Quotebacks[0] - snippets that are designed to be for all-purpose use rather than dev use.

To me, this is a possible sign that there is momentum back toward general Web blogging and home pages. While native apps have a lot of great tooling these days (like TikTok's "duets" or Snapchat filters), the Web excels at markup and embedding - and it's great to see tools that play to that strength.

[0] https://quotebacks.net

I'm afraid I felt that the spaced repetition questions on quantum country were a complete waste of time. There I was struggling (a lot!) with trying to work out how on Earth to think about C^2 (the space of 2-tuples of complex numbers, which you might semi-naively think of as 4-dimensional but that's probably a terrible mistake, but anyway that's not really germane to the point here) and the spaced repetition questions were asking me totally superficial questions, of the sort you might expect to be given to a 13 year old.

I couldn't help feeling that this was likely to be a manifestation of the common desire nowadays to make things "inclusive". I would love for things to be inclusive, but we mustn't let inclusiveness be correlated with dumbing down; that will in the end turn out to be a toxic correlation. And there is a limit to how inclusive you can be about complex vector spaces and their use in modeling information in such a way as to facilitate non-classical computing.

The homepage does not load if you use an adblocker that blocks Sentry.

Edit: Has been fixed in the latest commit.

Can confirm.

(Unless you're using an adblocker worth its salt, like uBO on Firefox, in which case it works perfectly.)

Reading this made me so incredibly nostalgic for the 2006-2007 era.

This is awesome man. Thanks for sharing!

Nuke them... from Orbit!

> Orbit is an experimental platform for publishing and engaging with small tasks repeatedly over time. In the short term, it's focused on supporting the "mnemonic medium", a way of augmenting texts so that readers can easily remember all the key details. For an example, see Quantum Country, a textbook on quantum computation. More abstractly, Orbit aspires to offer a more general form of spaced repetition systems like Anki, as part of a connected cloud service. Learn more on the home page.

Huh?

Reply on news.ycombinator.com