Current mainstream document (or long paper) translation is commonly done by converting the format, segmenting it, and then translating it, which requires specialized tools, for example:PDFMathTranslate ,GPT Academic Wait...
Of course you can submit the document as an attachment directly to the ChatGPT The smart spectrum, Kimi Waiting for the big model requires him to translate directly, with the disadvantage of not being able to preserve the original formatting in its entirety, which is not the point, but what is important is that the output of the big model is limited in length, and as the length of the output text increases, the degree of adherence to the initial translation instructions deteriorates (i.e., forgetting about the task at the beginning or what has been translated previously as the length of the translation increases).
Kimi is very good as the first big model in China that supports long contexts and outputs speed reading very quickly, and recently Google's Gemini Longer contexts are also supported, but in practice the output is still "short" when translating long texts.
When the translation is interrupted halfway (unfinished), we can type "continue" to let the big model continue to translate, if this method meets your needs, it ends here, Kimi and Gemini are both good.
Here's a better way to do it, which is to use the Conch AI The
Why use it, ref:MiniMax First open source MiniMax-01, 4M long text, new architecture, challenge Transformer , the most important feature of this model is that it supports 4M long contexts, and the most important point not emphasized is that the MiniMax-01 model can beRealistic output of extra-long contexts!
How long is the output? In practice, outputting tens of thousands of words is not a big problem, and theoretically outputting a million words is possible, and the reason for not reaching the theory is thatovertime payThe
If you use the Conch AI The web-based product for web page translation, which not only outputs you super-long translations, but can also convert theSave images in documents as online URLsThe translation of the entire document into another format is easy to do.
You may not have an idea of what I mean by outputting extra-long contexts, and end this with a screenshot of a translation task in progress: