Show HN: Never forget what you've learned (saveall.ai)
Save All uses quizzes, notifications, and emails to stop you forgetting what you've learned. Unlike other spaced repetition apps, we've (really) prioritised making Save All as simple as possible - minimalist design, no distractions, and minimal cognitive load placed on users.
We achieve this simplicity by using machine learning (e.g. large language models like BERT, GPT) to reduce the number of decisions users have to make. You don't have to decide whether you remembered a card, we know whether you did. You don't have to decide how to be quizzed on your information, we'll decide for you.
We're VC funded & growing fast but would love to hear HN's critical feedback. Tell it to us how it is!
11 comments
[ 9.2 ms ] story [ 60.2 ms ] threadSeems interesting, although Anki is still free and incredible.
Would be helpful if you had some discussion of the spacing algorithm you use.
I'm also curious whether the multiple-choice-ifying actually improves recall or not?
free recall vs. recognition
harder vs. easier
We use a similar spacing algorithm to Anki at the moment but eventually we will use a custom spacing algorithm per person & per card that uses machine learning to learn the optimum spacing algorithm per person & per card
Multiple choice questions are the least effective for memory but they help to keep people's motivation up and make the quizzing less taxing. We mixup the quizzes with multiple choice vs. type-in-the-answer so people are getting the right balance.
> In 1996, when Dolly the sheep was cloned from a single adult sheep cell, the world looked on in amazement. For some scientists, however, the most amazing thing about Dolly was the cell from which she had been cloned.
It appears that Save All cloze'd out 1996, but also "when Dolly the". This kind of unexpected behavior is familiar to me as my buddy and I built a similar NLP-based auto-cloze tool which gets tripped up trying to find important pieces of info to cloze out. In general, I'd say that clozing out dates is actually pretty low-value; personally, I (manually) clozed out these words in the passage:
> In 1996, when {{c1::Dolly}} the {{c2::sheep}} was {{c4::cloned}} from a single adult {{c2::sheep}} {{c3::cell}}, the world looked on in amazement. For some scientists, however, the most amazing thing about {{c1::Dolly}} was the {{c3::cell}} from which she had been {{c4::cloned}}.
Wishing you guys best of luck. When someone finally cracks this problem (generic content => nice flashcards), it's going to be huge.
And yeah the green words aren't very good for that Dolly example. It's ML based so its output is somewhat random but usually it produces green words that are much better than that so i think you were a bit unlucky with that example. All the data we collect on the clozes gets fed back into the algorithm so overtime it should become more consistent aswell!