Bro is living the dream. Got a legit 11/10 wife.

Base +24, jaingbing +6, CommonWriter +6

in response to the dude's paper

if someone is actually interested in using AI to understand a story you don't just feed it pages you use OCR to pull out the script and feed the script and then use the drawing context to correct the rare mistake or give it context on ambiguous wording. yeah people will rip on this too but I can read Japanese not horribly (my anki is at about 5000 cards, I've read some novels JP) and it does quite well like this since it has continuous speech. If the people pumping out the AI shit actually did this 99% of people wouldn't even notice.

Base +7

wow I didn't know there were tools to rip the text from images. As you say, that does solve the problem I saw from my testing where AI didn't have the full context and continuity of the story and actions.
And yeah, if people uploading AI took the time, only full-time translators would notice but the vast majority who are just hentai enjoyers will not notice.

Base +4

I can always appreciate someone doing things themselves imperfectly rather than just using a machine and passing off dogshit as hard work. I miss when the rough tl tag just meant someone did a mediocre to bad job. And, yeah, AI can produce a readable script - that's something LLMs are designed for - but there's no guarantee that it's accurate or even representative of the original work. At that point you may as well tag it as a rewrite instead of a bad translation. This is also the broader problem with LLM-based AI: the form is good, but there's no guarantee that the actual content isn't garbage. Why bother reading an AI "TL" if I can just make up porn dialogue in my head like I have been with untranslated stuff for decades?

Base +9, Shardshatter +6, CommonWriter +6

when you dump the script into chatgpt you can tell you to break it down line by line meaning by meaning and then you can easily verify its output yourself if you understand japanese at all and it's very accurate. It knows internet slang / manga cliche phrasing that you can verify yourself.

it's not fun or rewarding that these tools are actually really good now but they just are. The boogeyman that it's just totally inaccurate and super misrepresentative is not realistic. They were created by stealing and analyzing countless amount of human creation and now they're good. this isn't early DeepL days that literally was just random rewrites when it didn't understand something.

to be clear, most people don't take the effort of dumping the script into chatgpt to do a competent job. they're too lazy for that. the current thing people do of just slapping a page into a tool and using that is garbage.

Last edited on 13 August 2026, 18:03.

Base +7, Shardshatter -6, steveouniverse -8