FRESH
Hacker News
Hacker News Hacker NewsLet's take a real example to show how this works.
August 19, 2025. My wife called me in to help her decide what to do about a dentist that she thought was ripping her off. A couple of quick suggestions later, and she went to being mad at me about not having heard the problem through before trying to fix it badly. As soon as she was mad, I immediately connected with how stupid what I did was, and that this never goes well. But, of course, it was now too late.
Not a mistake I was going to make for a while. But, given my history, a mistake I was bound to make again.
I changed that. This time I stuck this into my spaced repetition system. Each time the prompt comes up, I remember that scene, holding in mind how it important it is to emotionally engage, not offer quick suggestions, and be sure to listen to the full problem in detail. It takes me less than 30 seconds. Reviewing this prompt, for my whole lifetime, will take less than 15 minutes of work. Just typing this up this time takes more work than I'll spend on it in the next several years.
This mistake hasn't happened since. Not once. And I believe it won't again in my life.
I have literally changed dozens of such behaviors. My wife says that it is like there is a whole new me. She can't believe the transformation.
All it took is looking at spaced repetition as general purpose structured reinforcement, and not as just a way to study flashcards.
Personally, I find the interface is extremely functional; the ability to have deck hierarchies to be a massive feature, not a bug; the WYSIWYG being the default being obvious given the intended audience, but one can still easily edit a textfile and import it or edit in html mode directly if desired; converting something into latex math is as simple as enclosing it in "[$] ... [/$]" and hardly the nightmare it's portrayed as; and finally potentially hacky plugins is a feature, not a bug: occasionally you have a very specific problem and some kind soul creates a solution for you, which may be functional but not the most aesthetically pleasing. That's fine. Anki is a bazaar, not a cathedral, and plugins have ratings and reviews which you can consult if necessary.
I have tried many different flashcard solutions, including hacky text-based ones, and I always return to Anki. Despite the fact that most other tools in my stack that I swear by are terminal-based.
1. You can view them anywhere (Github renders them nicely) 2. You can edit them in your favorite editor 3. Formatting doesn't decrease the readability 4. Extensible (syntax highlighting, mermaid, mathjax, etc.) 5. Cross-linking which is a core for any knowledge system is free 6. You can use Git for versioning and backup, etc, etc.
You could mark items in the feed to space repeat for yourself. This would also function as a “retweet”, which would align incentives such that content that gets promoted is actually durably useful or interesting. The posts people make would repeat to themselves too, so the source content should be good.
Wouldn't this invalidate card's review history if I am to fix a typo in the card's text?
Man I was really looking forward to seeing how they stored review history in plain text.
1. Turn the subject matter into a knowledge tree. 2. If a branch has more than 5 leaves, you split it up. 3. Flashcards are generated by traversing the tree. The parent node is the question, the child nodes are the answer.
The benefit of the tree is that it forces you to think about where in your structure a given piece of new information fits.
# Solar System
## Planets
### Color
- Earth ?:: Blue
- Mars ?:: Red
Path: Solar System > Planets > Color
Front: Earth
Back: Blue
The syntax also means that I can easily add cards from my regular Markdown notes, so regular notes and Anki cards live together.
1. The biggest win is just doing spaced repetition. Period
You don't even need an algorithm. You can just have options for "remind me in 1 day, 7 days, 14 days". This is how people did with physical cards: they just put the card at the back of the deck, the middle or the front.
2. LLMs now make it trivial to just say "make me an Anki clone in python with these features" and it will come up with something pretty decent.
In closing, learning the things that LLMs can't do quickly and efficiently is basically what we should all be doing.
I also often found myself wanting to make different flashcard decks from the same basic information (for Mandarin pinyin sentence --> character recognition, characters --> English translation).
If there was a sheets like data entry interface backed by a text format it would be great.l (I rolled things with streamlit but it's always cumbersome to get started).
A very simple cli tool, consuming basic txt format. You can use it in a second window while waiting for your compilation to finish.
Recently I’ve been also experimenting with defining QA pairs in my note files (in a special section). I then use a custom function in emacs to extract these pairs and push to a file as well as Anki.
The biggest area for improvement is probably deck collaboration. Most SRS proponents often state that its bets to make cards yourself because the act of making the cards is a key part of the learning process. I don't disagree, but part of the reason that making cards your self is recommended is because the shared decks are, on average, terrible.
After that I would like to see more built in support for non front/back or cloze cards. There are a lot of other card types that you can make, but are difficult or impractical to do in anki. Things like "slow" cards, one sided cards, code/music/math/text cards. These can all be done in anki, but it's a pain.
Then support for card order/hierarchy/prerequisite an and encompassing graphs like what MathAcademy does.
And lastly, a web first experience. Anki is offline/local first. That has the benefit that you are always safe from being rug pulled. But there are a lot of places (like work) where local first does not work well.
Every product will eventually use markdown as their content store.
Could you imagine adding support for this?
I have use anki, and briefly mochi.
Having plain text cards that are simple to edit and manage with basic linux tools is really important.
I have used the genanki python library in the past to generate cards, but it's not great.
Going to give this a go.
The data format of Anki is a bit complicated but at least it's SQLite. I've seen a ton of shared decks and resources on ankiweb, but it's true you can't easily put them on GitHub.
I've thought about posting to HN but I'm a little apprehensive of when and how to post.
Anyone interested in this and/or have some advice for posting my prototype online for feedback?
You can customise note types with CSS and Javascript, which means that you can make cards look however you want.
If the cards are identified in the database as their hashes, wouldn’t editing the content reset all repetition data so far?
Anyone here has been using FSRS long enough to have comments about its effectiveness? I think it’s general consensus that moving from SM-2 to FSRS will show great improvement. I’m using SuperMemo 9 though, so it’s much harder to understand whether there will be an improvement or not.
1. Decide on what's important. Just because you learn something doesn't mean that it should be logged to the system. I used to log a lot of minor details (like niche method signatures or command flags to the system). If you make cards for every detail like this then you will be trapped reviewing 100s of cards daily that you likely never use.
2. For the cards you deem are important, make sure you understand the concept. This often means making 2-5 cards for the concept that test your understanding from different angles (definition, pros, cons, how would I explain this to someone else, etc...). This helps to cement the concept at a foundational level.
3. Try to move from the existing flashcards to 2nd order flashcards or pure application after the first couple reviews. So your foundational cards are now set to review in 6 months or 1 year. At this timescale if you prioritized what was important and made sure that you understood the foundational concepts, then usually simply doing things related to the concepts will be the reviews (and sorry to say but if in 1 year you get a card related to what you are doing, but never used, chances are it probably wasn't that important). In addition to doing, you can also create 2nd order flashcards (which might compare 2 concepts). These types of cards test the foundational knowledge indirectly, and are helpful for higher order thinking.
In conclusion, I think spaced repetition is a very effective tool for efficient learning (especially in the first 60 days or so after learning something). I think the major pitfall is not prioritizing what cards get made and being stuck in review hell.
Couldn't agree more. I think I would take this opinion and go even further -- we shouldn't be making cards fully by hand much, if at all, anymore. AI-assisted card creation is to me clearly the future, and already AIs are good enough for this to work well.
I think uploading a textbook to NotebookLM and getting it to build out a deck for each chapter will be a great study method.
I'm happy to hear other suggestions too?
The decks for studying Japanese that I’d like would have RTK/wanikani style elements used for mnemonics and I’d like them shown in the answer along with a full description and cross references.
Right now I’d have to build a templating system to prebuilt my deck and import it and it’s just a lot of work on top of the work of building the content, but mostly it makes it difficult to edit/update cards while studying.
> Your performance and review history is stored in an SQLite database in the same directory as the cards.
org-drill is the original main package, but the newer org-srs is probably better (and supports FSRS).
Do you use Syncthing or something else to sync your performance history between devices?
Any way to use them, or do I have to go through markdown format?
It has the least friction for creating flashcards I’ve ever seen. You actually don’t even have to create flashcards - you can add any note to the review queue with one keystroke and record the ease of recall with another command.
With regards to cloze deletions the author writes:
> cloze deletions in Mochi are very verbose. [...] This is a lot of typing.
First, the numbering (1::) is optional. Secondly there are keyboard shortcuts, cmd+L to wrap in {{}} and cmd+1,2,3 to add numbering.
The point about note types is fair, and I may a similar function eventually, but I recommend most people to create no more than 10 cards a day. Any more and you risk getting overwhelmed with reviews. In the article the author shows an example of creating 4 (or more) cards for a single atomic element. This excessive card creation probably contributed more to the 1700 overdue cards than the algorithm (more on that later). If you really do want to create multiple cards like this you can use cloze groups. E.g. {{1::Helium}} (symbol: {{2::He}}) has atomic number {{3::2}}
Finally, the "biggest problem with Mochi". This is kind of a moot point now that Mochi has an FSRS option, but there are a lot of misconceptions in the article about the algorithm. First being that Mochi's algorithm is inferior to SM-2 because it is simpler, and that the rational for it being simpler is because "the user can reason about the algorithm more easily." I'm not sure where the author got that idea, maybe I mentioned it before as an advantage, but it's not the main reason. The main reason is that the additional complexity in SM-2 is actually detrimental in some subtle ways. [0] The author just assumed the algorithm was worse and gave up.
With regards to the forgetting multiplier the author states:
> If I forgot something after sixty days, I surely won’t have better recall in thirty.
But what is the evidence for this? The assumption here is that the knowledge was "completely lost". For the card to have gotten to 60 days in the first place, you must have remembered it previously after 30 days. Evidence show that reviews strengthen memory, not degrade it. Even FSRS does not completely reset the interval after a forget. I get that the author doesn't want to configure things, but lowering the multiplier to 0.2 for example seems a lot easier than building a brand new SRS flashcards app.
Criticisms aside I really do like the idea of hashcards. Plain text, offline, open source. It checks a lot of boxes that I personally look for in software and I'm happy to see more options in this space.
[0] https://web.archive.org/web/20200926103540/https://massimmer...