As I’ve been learning Spanish these past months, I am almost compelled to create an LLM-powered language learning application. It is really hard to do spaced repetition without it turning into a grind, and the ability to create (and parse!) custom, idiomatic text programmatically could be an absolute game changer here.
They are also really good at answering questions about how language use and I haven’t noticed much (anything?) in the way of hallucination with frontier models.
@pganssle Same for me, Japanese SRS keeps becoming more of a grind, there's so much to memorize and without more context it can get really boring. I keep wanting sentences I can read with what I know so far, a lot of the time it's a sea of unfamiliar vocab and grammar. My next step is going to the public library, they have quite a few children's books.
@davidism Yeah, you may want to look into comprehensible input. I’ve been using it to learn Spanish and among the 6 foreign languages I’ve spent time learning, this is the one I’ve made the most progress the fastest.‡
See: https://comprehensibleinputwiki.org/wiki/Main_Page#Japanese
‡This effect may be confounded by the fact that I learned 5 other languages first, evidence for which being that I was able to have simple conversations with Spanish speakers even before I started learning the language at all…
@pganssle Excellent link, thanks!
Of course, a countervailing force here is that they are also stupidly good at translation, so the utility of actually learning another language is reduced.
They can explain nuances in a way that automatic sentence-to-sentence translation can’t, and you can give them enough context to let them know how to select the right way to translate what you want to say into idiomatic speech.
Of course, you will sound like an obsequious PR person if you don’t take steps to avoid that. 😛