Bringing AI into Bunpro

Bunpro
#1

To put my opinion out there I am a fan of bringing AI into learning Japanese, specifically into Bunpro. I’ve recently seen that some people are against the idea for many reasons that I can understand but I just don’t find them convincing enough to not implement AI. I’ve seen a lot of really bad implementations of AI into language learning, but I’ve also personally used it for a lot of really good ones as well. I think it’s a technology that has a lot of potential for making Language Learning much better and it would be kind of bad for the long term if Bunpro decided not to use it.

When I reference implementing it, I am not talking about making AI the focal point of the site. Rather including it in for further clarification, potential for guided learning paths, creating practice sets for user specific struggles, reading and writing practice, and many other uses. With the continual advancements that are being made in LLMs I think more of the issues with accuracy and consistency will continue to go away.

I just wanted to start a discussion about it to see what others think.

1 Like
#2

AI is already being forced into everything else I already use.

Where does Bunpro need it? It is a complete product as it is imho
Would you bake AI into Anki?

21 Likes
#3

In the FSRS discussion they were referencing different ways to do spaced repetition, and it seems like if you implement AI correctly you can actually see really good results that beat just about anything else available. If that was possible on Anki then I wouldn’t mind them adding it at all. Imagine something like that for Bunpro where the review windows are more tailored to your learning.

I’m not saying make Bunpro the AI language learner but there are plenty of ways that AI can help with improving the speed and enjoyment of learning a language.

#4

I am strongly against it. AI is a flawed tool that is already being overused as it is. If people really feel they need to consult AI about something, they can go to one of the big AIs and ask them a question in their own little box. There is no reason to bring it here.

19 Likes
#5

I feel like a lot of the issues with AI are on the implementation of it more than AI itself. A lot of the best uses for AI are things that aren’t front facing to begin with, features like scheduling cards, reccomended areas for extra study, book, manga, and TV reccomendations based on the words and grammar you specifically know. I think a feature that could take all of my information already in bunpro and a category and find me media to practice consuming that I could enjoy and would match my level would be great. It would save many people hours of searching and trying things out!

2 Likes
#6

You could argue however that trying out new material, and finding out for yourself if it’s appropriate for your level - whether too easy and you breeze through it, or too hard which you try, get stuck, and come back later to it - IS part of the fun, and the learning experience.

Optimizing according to your data takes that experience away too.

5 Likes
#7

That’s a fair point. There is a huge sense of reward when you can go back to something you wanted to read or watch before but were unable and now are. I would just add two things:

  1. I think people will still try to find their own things because everyone has shows or books or other media that interest them and possibly influenced them to start learning Japanese.

  2. There are people who are anti Anki and other SRS because it’s not the proper learning experience and how they did it and it’s better to naturally absorb words or to start with reading and to remember that way.

From my point of view Bunpro was made to help people learn Japanese in the best way they could. There was a big gap in the market for people making grammar resources as great as they have and I think they’re still working very hard to keep providing that. I think that AI is just another good way for them to keep doing it.

#8

What’s stopping you to do that in an AI of your choice on the side?

I am against any implementarion for all the argument other users stated above but one I rarely see is the argument about cost.

If you use AI you have to pay for the token usage and Bunpro will be locked using the AI API they choose and at the mercy of the price increase.

How so? Everything the AI will write would need to be reviewed by a human as the grammar here is so specific that the AI will mix things up so not a lot of time gains. Then why use AI at all?

For recommandation, there is already dozen of list online and on reddit so AI will nicely summarize it for you as it (sadly) loves reddit and feeds on it.

We don’t have to have AI everywhere, let us have Bunpro AI free please.

I wouldn’t only slightly agree on a algo change for the current SRS but that ciuld be done without so AI so why implement it there?

4 Likes
#9

Bunpro already uses gen AI for vocal placeholders and that’s more than enough. It’s one of the few learning apps out there that hasn’t fallen prey to the AI craze and I’d want it to be kept that way. I paid a lifetime for an app that’s made by Japanese-loving people, with language learning experience, not AI-garbage.

8 Likes
#10

Why not do it myself outside of Bunpro:

The ease of access that Bunpro already has to all of my learning data is what I think would make it much easier for them to implement AI that’s centered around the user. You could go and try to gather all of that data yourself and then compile it into a format to use but it makes much more sense to do it in Bunpro. They also have access to a very good and organized structure for learning that is easier to incorporate into building around what and how people what to learn.

On Cost:

It is not that expensive at all when you implement AI effectively to provide a lot of these features. For reference using OpenAI API Compare models | OpenAI API

It costs cents to send and receive hundreds of thousands of words.

Switching:

When it comes to switching you aren’t really locked in at all. You keep all of your backend logic and prompts and just choose whichever model and provider you want at anytime. If you don’t like one you can go to another as easily as choosing and getting your api key.

AI Being Reviewed:

That’s assuming you are making generated AI content in Japanese. In that case you probably will want to review it, but I can almost guarantee it will be much faster to make an AI generated story and have someone review it for grammar and vocab mistakes than to write one from scratch. They also never have to make AI generated Japanese content if they don’t want to there are still plenty of ways to use AI to add features and things to help everyone learn.

Outside Resources:

There are tons of resources out there for people to go look at. The thing is it takes a lot of time, isn’t specific to a person in the way it could be and becomes quickly outdated as new things are released. None of this is needed just like you could argue all the grammar on Bunpro is online for free now and you can make your own grammar SRS with anki.

AI isn’t an evil force that ruins companies and products. People that choose to make trash and feed it to their customers are the ones who do that. AI is just a tool that can be really amazing at helping people learn and that’s why I am for it.

1 Like
#11

There’s a world of difference between smart implementation of AI for things like FSRS calculation and other machine learning contexts, and stuff like chatbots and genAI. And all of those being constantly spoken about and sold under one big “AI” umbrella term really does a disservice to any suggestion or discussion.
The slighest mention of “AI” and it already gives people the idea that you’re suggesting putting in a chatbot into Bunpro or something, even though you might just be suggesting machine learning that predates chatbots by years (cf. for example frogg’s reply about AI-generated audio already being more than enough and not wanting any more AI garbage)

With that said I’m against the implementation of AI generated content, however if it’s AI application for things like FSRS calculation and algorithms I wouldn’t be against it.

6 Likes
#12

this can be done fully deterministic in a much easier way than using AI for it. You are looking at a score-based vocab knowledge overlap matching algorithm rather than LLM-blackmagic

4 Likes
#13

This feels like the equivalent of adding labubus or fidget spinners to bunpro

I say it lovingly

I personally don’t think bunpro needs much more to it tbh, and nothing AI could bring

There is a myriad reasons why I could say AI is better out of bunpro than in it, but the biggest thing to me is that I don’t think Bunpro needs to be all things to all people

If you want or need to use AI then there’s services for that already

8 Likes
#14

1000011104
This topic is divisive haha.

Thank you for taking the time to answer :pray:

2 Likes
#15

Well you also don’t need to use an LLM to implement AI at all there are many many smaller models that can be run locally for specific tasks I was just generalizing for the sake of discussion but yes that makes more sense than an LLM!

#16

Boooo, wish there was a way to dislike or down vote you. AI = BAD

4 Likes
#17

point being it’s a very simple algorithm, it’s not a use-case for AI. Wanting to use AI for it is part of a larger, unhealthy trend where people/companies look at the easy way to do something half-assed and pick that rather than spending the time to write an appropriate solution instead.

a very simple, deterministic score-based matcher will outperform every single existing AI on this task. I say this confidently, as someone who has been in Software Engineering for the past 16 years.

5 Likes
#18

Also ngl im kinda worried about the AI bubble bursting at some point and I worry that if we integrate AI into an already perfectly serviceable website it will cause problems down the road

9 Likes
#19

Thank you too!

1 Like
#20

I kinda referenced it earlier but there are many locally hostable AI models that are very small and will never be affected by an AI bubble burst. They still are “AI” but you can think of them as fancy algorithms for specific tasks that you host on your machine and aren’t tied to any outside companies so there is not risk of the site closing due to AI crisis!

1 Like