i do actually have one application of LLMs as an anime translator, even though they’re otherwise almost completely useless in my line of work!

it’s a rare one, though: figuring out how to transcribe non-english/german names or terminology the japanese mangled into katakana and didn’t bother to include in their scripts and supplemental name lists. these tend to be so ambiguous that guesswork is all i have, and there’s not really any way i can ask the production staff or author of the original work and get an answer in a reasonable time frame (or at all).

so i’d query a model for possible transcriptions given what i know about the person—gender, nationality, culture, etc.—and save myself hours of search engine queries (assuming i have time for that in the first place) if i’m lucky. if necessary i can narrow it down like, “this character is from that specific region and born between 198x and 199x” and get statistically likely candidates, training dataset permitting.

this is what those models are actually quite good for. not great because they tend to confuse given names and family names for example, and obviously make shit up when the katakana version is really just gibberish meant to sound exotic, but it’s a nice tool to have that has at least once helped me get something more plausible than i would have managed on my own

@mia Wasn't this what AI really been. A tool in your bag making work easier and faster not replace you?

@stefan in an ideal world.

in THIS world, however, the bottom liners are always trying to cut costs. it doesn’t stop them from using machine transcription and machine translation and then finding some freelancer who’s having trouble making ends meet and can’t be picky about the jobs they take, and then pay them a fraction of what their work is worth for basically redoing everything from scratch because if the customer complains they won’t even see THAT money

@mia sad reality is sad