What algorithm are you using for SRS? Anki recently updated to include (the much better) FSRS4.5 algorithm alongside their usual (and still default) SM-2.
In-case you haven't updated: FSRS 5.0 was released in Anki 24.11
Show HN: Ricotta – Language Learning to Replace Anki
21–30 of 83 posts
Re: Show HN: Ricotta – Language Learning to Replace Anki
#22I'd have to see some serious stats to believe whatever SRS algorithm you use is better than FSRS (which is supported in Anki). Stats for FSRS here (and elsewhere) [0]. [0] https://www.reddit.com/r/Anki/comments/18csuer/fsrs_is_now_t...
Yeah as someone working on educational software, anything hand-rolling their own SRS is a pretty big red flag. Beating FSRS is going to be next to impossible, especially FSRS with parameters optimized from your users’ review history.
Re: Show HN: Ricotta – Language Learning to Replace Anki
#23I tried a little Japanese (the second language I know best). Not sure where the content comes from but the translations were wrong, the vocabulary was strange, and otherwise the cards did not work out for me. Seems like they were auto generated via AI or google translate or something, in any case they were not accurate or natural sounding. Maybe other languages are better but I wouldn't know enough to tell.
Anki is one of my favorite pieces of software ever so I was definitely willing to try yours. Seems like the feature that defines your app is auto-generation of a card / set of cards. Didn't see any way to create your own card, all you can do is put in an English word and get the automatically generated card added to your deck. So if that's not working well the entire thing crumbles.
Interesting idea that might have a future if it finds significant improvement. I think if it was ironed out I would still prefer it to spit out an anki compatible deck that I can use in Anki itself instead of in the ricotta web app.
Re: Show HN: Ricotta – Language Learning to Replace Anki
#24I tried a little Japanese (the second language I know best). Not sure where the content comes from but the translations were wrong, the vocabulary was strange, and otherwise the cards did not work out for me. Seems like they were auto generated via AI or google translate or something, in any case they were not accurate or natural sounding. Maybe other languages are better but I wouldn't know enough to tell. Anki is o…
If you have iOS/macOS, I've made a native app for mining Japanese web/ebook content into Anki (or Manabi Flashcards which is getting FSRS soon too): https://reader.manabi.io
I'm currently working on adding first-class Mokuro integration for manga mode, bilingual user-provided or Whisper-generated captions for video, an HDMI-input mode with realtime lookups, and local/BYOK AI features. And once I add Yomichan dictionary support, I will be able to go multilingual too. I also have a beta available via the Discord that resolves some bugs that I'm trying to release very shortly.
I've added Anki integration and am working on WaniKani + JPDB sync. Hopefully more service integrations soon too. I like to have nice defaults but let users keep their flashcards wherever they like. Ricotta too if there were an API.
Re: Show HN: Ricotta – Language Learning to Replace Anki
#25Congrats on the launch!
Minor nitpicks:
Dark grey text on black background (navigational menu items, testimonials) is not readable.
Section 5 of your Terms and Conditions are, in part, utter nonsense. It says there that as a user of the service, I have to display cookie banners (how? where?) and other duties that make no sense whatsoever. Are you sure that this is what you wanted to say?
Re: Show HN: Ricotta – Language Learning to Replace Anki
#26Just tried it for German. The cards gave me the word without their article. Because grammatical gender is so important and impacts so many aspects (adjectives, case markers, etc), you really should learn the word with the article, otherwise you'll be in a world of pain later.
Re: Show HN: Ricotta – Language Learning to Replace Anki
#27People stop naming things after real existing words.
Re: Show HN: Ricotta – Language Learning to Replace Anki
#28It doesn't handle homonyms well, since the AI isn't given contextual reference. The sentences are also too simple on advanced level, so you don't get much more out of it than just reviewing the words. The AI coming up with words for you doesn't add a ton of value, most languages have standardized tests with good word lists that learners should already be using. Adding the issues others mentioned, I think this is feasible but needs agents taking additional steps to mimic what a real tutor would do and formulate a useful short lesson.
Personally I find Duolingo great for Japanese, I have the opposite problem: I run out of hearts if I try to do review and the app rarely gives me much old content, so I don't practice it. If this could come up with Duolingo-esque lessons for different levels I might use it to supplement. But letting users choose words/a topic is not necessarily important, most people don't know what they should study so it might be better to have "recommended" premade decks for different levels.
Re: Show HN: Ricotta – Language Learning to Replace Anki
#29I'd have to see some serious stats to believe whatever SRS algorithm you use is better than FSRS (which is supported in Anki). Stats for FSRS here (and elsewhere) [0]. [0] https://www.reddit.com/r/Anki/comments/18csuer/fsrs_is_now_t...
not OP. I have a competing product with an ad-hoc algorithm that I 100% believe is better than FSRS (tested it for months or real practice)
But what you ask is not feasible, neither I nor OP have 700 millions of reviews to stack up against FSRS.
I have my own experience and a bunch of simulations, and if FSRS were better I would have switched to it.
The reason in my case btw is that FSRS is generic, mine is tailor made for languages - big difference.
Re: Show HN: Ricotta – Language Learning to Replace Anki
#30I'd have to see some serious stats to believe whatever SRS algorithm you use is better than FSRS (which is supported in Anki). Stats for FSRS here (and elsewhere) [0]. [0] https://www.reddit.com/r/Anki/comments/18csuer/fsrs_is_now_t...
Yeah as someone working on educational software, anything hand-rolling their own SRS is a pretty big red flag. Beating FSRS is going to be next to impossible, especially FSRS with parameters optimized from your users’ review history.
FSRS is so cargo-culted. It's just an algorithm, claiming that any algorithm can't be improved is ridiculous.