2 comments

[ 3.9 ms ] story [ 11.7 ms ] thread
Update: Google translate already fixed the issue, so it's not reprod-able.

I find this is very interesting. It does not look like a bug. To me it looks like someone hacked the translation NN model for a particular pair of input/output?

Twitter labeled the pictures as "potentially sensitive content", despite it being just screenshots of the Google Translate text boxes. How on Earth does Twitter make these decisions?