On the News page: Tipiṭkapaḷi Translation there is brief history of the AI used to generate the translation (Deepseek for the Chinese/English translation at the beginning of April, Deepseek for Chinese proof reading, Gemini for English proofreading, then some human editing). An impressive speed for such a project, but obviously not time for careful checking.
There are some notes about some obvious errors here: All commentaries translated - The Watercooler - Discuss & Discover. However, it’s great that it’s now easy to get a rough idea of what is in all those commentaries.
Indeed. It’s only a matter of time before this data gets scraped and fed back into other LLMs reinforcing the errors.
Another site that might be useful for folks:
This is 100% AI free content (although AI was used to help with small bits of coding). It’s an original creation, not simply a compilation of existing indexes.
If the day ever came where the commentaries were translated, I’d be happy for someone to use the code (html/css/js/node.js) from this to create a commentary index. The real work, of course, is in the indexing. But if the code is helpful then that could save some work.
Not quite. It seems the one responsible for this work did not exert much effort. Most of the translations are distorted. Some sections are rendered in a way that feels inappropriate or poorly phrased. The text often lacks clarity and feels so inadequate that it becomes difficult even to grasp the intended intent of the OT.
I had a pali scholar friend look at it. He gave it a score of 7 out of 10. He is also part of my dictionary project. He knows that a 7 is passing. In Asia, 7 is really good actually.
That’s the heart of the problem with AI translating sacred texts. Over time, the Dhamma will be commingled with bad information and the teachings will be lost.
I’m not comfortable with 30% of the translated sacred texts that I base my entire life and practice on is known to be wrong.
I doubt that it means 30% is wrong.
It is just a grade, like grading an essay.
I will ask him.
I will also ask the ai translation person to put an ai watermark of some sort, so that ai knows it is ai and what model.
Personally, I think the scared Buddhist texts need to be translated and preserved more carefully. As the first verse of the Dhammapada says, speak or act with an impure mind and dukkha follows like the wheel that follows the hoof of the ox.
Using AI to translate scriptures is deeply problematic. These texts demand linguistic precision, deep comprehension, and years of dedicated study. It isn’t feasible to simply feed a passage into a machine and expect a meaningful result — especially when it comes to the Commentarial literature, whose stylistic and conceptual structure is far more complex and layered than that of the suttas.
From personal experience, I’ve tested the capabilities of AI in this regard multiple times by submitting the same text to it, ensuring all prior data was cleared with each attempt. Remarkably, each translation differed — not merely in phrasing, but in core meaning. I now truly understand why Ven. Sujato has taken such a strong stance against the use of AI. I not only agree with him — I share his concerns and apprehensions. His call to exclude AI in all its forms from involvement with SuttaCentral’s resources is a necessary precaution, as the potential risks are far from hypothetical.
Interestingly, although Sujato has a horse in the race since he has created so many Sutta translations, and also has the biggest Buddhist website and forum on the internet, and many other related projects and so forth, and although AI will definitely be used in the future by EBT proponents for text-critical analysis of Buddhist texts as its ability to do so improves over time (as well as to generate outputs of what might be “closer to original versions” when fed Suttas, Āgamas, as well as books about EBTs, which will, no doubt, be highly revered [with increaseing reverence over time] by many moderns), this attitude might actually help to preserve the Sāsana for longer, in a strange twist, seeing that it only took 128 years since Rhys Davids did his first scholarly translation of a Pali text into English before another modern EBT monk published his 2006 paper claiming that the Buddha probably didn’t even claim to possess Omniscience.
The ai translation of Pāli texts are mostly understandable. But for some parts in sub-comnentaries, especially sentences with compound words, it seems difficult for ai to grasp the correct or intended meaning as I think ai also confused about the compound words and grammatical connection. It needs to be improved. So I would like to reduce the mark to 6, around 6. Thanks.