News for hackers. Stuff that matters.

A comment on Almost fifty thousand roman coins discovered in the waters of Sardinia. Parent comment

Re:Almost fifty thousand roman coins discovered in the water... by batch12 on Tuesday November 07, 2023 @01:56PM (#38181166)
I'm seeing indications that this is sometimes related to language translation using GTranslate.io [1]

> We use XNUMX variable instead of all numbers in our edit interface. For example "10 cents" in your original text is represented as "XNUMX cents" in the edit interface, so if you translate "XNUMX cents" once, the same translation will be used for all similar phrases like "11 cents", "18 cents", etc. This helps to do less work.

[1] https://docs.gtranslate.io/en/articles/1348921-how-to-edit-t...

[ Parent | on HN ]
Re:Almost fifty thousand roman coins discovered in the water... by manojlds on Tuesday November 07, 2023 @02:29PM (#38181594)
Tokenization like this is why LLMs are also bad at math.
[ Parent | on HN ]
Re:Almost fifty thousand roman coins discovered in the water... by eevilspock on Tuesday November 07, 2023 @02:09PM (#38181340)
That explanation tracks with the number of non-English results in this search: https://www.google.com/search?q=xnumx+-michigan&tbs=li:1 and the way XNUMX appears in these results.

(apologies. I only use google when I need to do a verbatim search)

[ Parent | on HN ]