When I ask Meta Ai to translate text from page to English, the voice translation is different from what is displayed in the translated page... like completely different. The voice translation is more accurate but the text translation is a sloppy mess. Why can't the text translation be from the voice translation?
Ahh yes.
That's because the text isn't the transcript of that what Meta AI says. When Meta AI "sees" the text, the context is more clear and the translation more accurate. I guess the shown text is the extraction of the picture which is then translated contextlessly.
That might be some sort of safety measure or energy saving thing, because live caption of Meta AI and Meta AI speaking would consume too much energy... I am not sure.
I agree with you that the voice translation is indeed much easier to understand.
Gemini handles the text translation pretty well. I've compared what Meta AI sees vs what Gemini sees, and there's a huge difference. Meta just blindly translates the words without context (just a translator from 2010) while Gemini is smart about context and applies real AI to it. It needs some serious update on how text is presented. It should done post voice precessing.
