[LgLasagnaModel]

“Almost perfectly”

It is freaking amazing that this works almost perfectly. Seriously, it’s mind blowing. The problem is, when you keep in mind how the technology works, you realize that the “almost” can never be removed. That’s fine for some use cases but not for others. I understand that human translators make mistakes but they have a conception of truth and correctness, that matters.

[brigadier132]

We have some people that read Mandarin and double check the output once in a while. If it didn't work well the story would quickly become incoherent and make no logical sense because chapters are translated on their own.

The common failure mode is names and genders, for some reason it likes to swap names and genders of characters.

[LgLasagnaModel]

1. I believe you that it works well

2. Plausible != Correct - ie, would someone notice if they were reading a slightly different but still logical story?

[brigadier132]

My point with regards to 2 is that it would have to maintain consistency across translation runs. Entire novels don't fit in the context so it can't make up a logically consistent novel across prompts.

When I do the translations, I actually don't even include previous chapters in the context.

[Ekaros]

So the last novel I completed is long one, but not unheard of I think it had 6 million characters. Now I don't know how many tokens would that be, but I doubt most models can support that large context.

And really consistent editing and choices between what is translated and what is not and instead is Romanized is rather important with many Chinese novels.

You can get something you can figure out, but I doubt you get something really enjoyed.