Isaacdl, this is an amazing project, with a lot of potential among non Chinese TT audiences. I'm not an expert but I collaborate with neuronal translation experts and I know the kind of difficulties you are facing.
Machine translation has improved dramatically in the last few years and it is already very efficient even with Chinese texts. However, to train AI to translate specialized texts, a bilingual corpus of hundreds of thousands of segments is needed, and I guess that that is not easy to find for table tennis. Another problem is that the Chinese language spoken in this kind of videos is far from being standard, not only because it belongs to an informal register but also because the players and coaches who speak there are not high profile speakers --let's say, university professors or TV anchors-- and they improvise a lot in a very unstructured way and even use expressions coming from other Chinese languages that can be easily understood from the context by native speakers but at the same time are a huge hazard for machine translation.
An example. I'm not a native Chinese speaker, but I guess (please, those of you who are better positioned correct me if I'm wrong) that the expression 十板球 and 六板球的能力 that Fang Bo uses in 1:20 is an influence from his local language. It has nothing to do with ten boards (even when 板 means board) or being able to loop ten times in a row, as the IA interpreted, but here it just means that Fan Zhendong's backhand is perfect (10/10) and his own backhand is only fair (6/10). To overcome this type of difficulty, the AI needs hundreds of similar texts, the bigger the corpus the better, and this is something that we currently don't have.
So it is not only a problem of the thousands of Chinese characters currently used and the new terms created by combining them in specialized jargon like that of table tennis, but also of the scarcity of bilingual informal texts to train the AI to cope with this kind of video.
In any case, here there are a lot of Chinese speakers who can contribute to fixing this kind of mistake and help to training the machine. Even at this stage, and even with the errors yet to be fixed, it is already a very useful initiative.
Here is my modest contribution with the first error, "quot;? It's all river noodles": "it's all the red face/rubber".
BTW, isaacdl, it would be useful to include the exact minute and second where the error appears on the video to locate it faster.