掲示板 Forums - renshuu's stance on AI

(This has been mentioned before in the renshuu news, but it was requested that I post it here as well for a permanent place to see it.)

I originally made renshuu to help myself learn Japanese, with the ultimate goal of using the language to talk and interact with people.

Aside from all the technical, quality, and ethical issues of using AI (most of which I agree are serious problems), our core belief is that language learning should be focused around people and the ability to interact with other people during your learning journey.

renshuu does not use any AI (LLM) generated user-facing content, nor will it. This means no audio, artwork, or text made by AI.

All artwork is done by Saki (my wife, and partner in renshuu) or members of our learning community.

All audio is recorded by native speakers.

All text (lessons, sentences, etc.) is written by people.

All translations that are not taken from publicly available dictionaries such as jmdict are written and verified by members of our learning community.

If you ever feel that you see something written by AI (say, a user-submitted comment or a user-submitted image for a vocab word), just let us know so we can look into it!

Disclaimer: I am the only developer/programmer for renshuu. I've been writing code for 30 years, and love it, but I will occasionally use bits of code that I generate together with an LLM. Even though you cannot see most of the code, though, 99.99% of it is written using the keyboard in front of me, and that will continue to be the case!

Of course it's still the case. You should recognize the voice as the male voice that has done the audio for thousands of the sentences over the last few years.

Edit: I'll go ahead and note that the voice artist I use it fairly prolific in his work, and has what I'd call (in a good way) a someone stereotypical voice you often hear with certain characters in voiceovers/anime. It's sad but understandable that a voice like his would become so prevalent in AI recordings.

It would have seemed that way because up until now, those words had no audio (and so their listening mastery was at 0). When the audio comes in, there is suddenly some new ones to train up from zero.

The おばけ one *did* have a weird "blip" right in the middle - I neglected to catch it, but did remove it for now.

I would like to make the case for AI. I think there is a lot to be gained from the new technology. Perhaps this could be done while still keeping Renshuu a place which values people.

If it were me, I would have all the word definitions re-written by AI. Some people would say this is extreme. But every time I look up a word in Claude, or Chat GPT, the LLM's explanation is better. I think it will surprise people if they haven't been using it this way.

The problem with the existing dictionaries, including Renshuus are:

1) they're inconsistent. The formatting of words is different, some have a single word, some have a whole paragraph, they have obviously been defined by different people with different goals in mind.

2) Many (most?) words with multiple definitions are ordered by historical relevance. For example, 投げ出す, now, usually means, "to give up on". But the first definition in Renshuu is "to throw down, to throw out, to stretch out (one's legs)". What we new language learners really need is a dictionary which lists the most common use of the word first.

3) The meanings or definitions themselves are out of date in many cases. e.g. 習い事. I have been learning this for years to mean "accomplishments, lessons (in an art, skill, etc.)". That's Renshuu's current definition. It never made sense to me. An accomplishment is something you get after finishing the lessons. How can a word mean both? I finally asked Claude. The reason is "accomplishments" refers to the archaic use - that one's extracurricular activities, like sewing or calligraphy, used to be called "accomplishments". So it really does just mean extracurricular lessons. Language changes over time. Ideally we would update word meanings.

4) There's not really another good method to update word definitions. I see that there is a box which says, "Submit to dictionary for approval" when I change my word definition, but are these actually checked? I have occasionally submitted definitions for words, and I see that they have not been updated. And then I think, "why would anyone trust me?" Every single person who submits is just one person saying that they can do better than the last. How can you get someone impartial to judge?

The great thing about AI is that the computer can read thousands of examples of word's use, and distill them.

Regarding your use of it, マイコー, in programming - I hope that you're not feeling guilty. It's fine to use AI to help with code. If it makes the site work well then why choose a slower route? Artisan code is not going to give you more time with your loved ones.

AI is just another tool, after all.

pup72 (10月27日, 14:24)
What if Bog Playmate is AI himself/herself/itself though?

Maybe I should have left the spelling mistakes in!

I appreciate you taking the time to leave your thoughts on it, but that's the great thing about the internet - each person can choose a tool(s) that align with their values, beliefs, methodologies - whatever.

As to your specific question: the "submit for dictionary approval" is mainly for non-English definitions. For the corrections to the core dictionary data, there is a "Report mistake" link that is checked (although not that frequently). They are never simply accepted blindly, but we take the time to do some research on it before approving/denying it.

I do believe that renshuu could be clearer on the pathway for that particular "custom definition" link, and I hope to get to that at some point.

As far as the code / llm stuff goes: I do understand your viewpoint (and I have family members who feel similarly), but that viewpoint reduces coding to a single pipeline - output. I like programming, I like the development, I like the puzzle solving, I like the sense of pride when I make improvements, figure out something clever, etc. I like being able to disconnect from the rest of the world and spend a few hours diving into code. I do not believe that I have some kind of imperative to prioritize output above my own feelings and my own experience. We have numerous other reasons as well, but this is one of the main reasons I have never tried to make renshuu a "proper" company, with employees, growth as a primary goal - I don't want to have to balance all of these other metrics with the things that make me and my wife feel good when we work on renshuu.

I realize that this is not how everyone feels, and I believe that everyone who develops gets the right to choose how they approach developing, what matters to them, and that's great!