Mastodawn

@randahl

An addendum: Youtube likely uses Google's own Google Translator under the hood, which in turn have been using Google Gemini under the hood, which in turn "glitches" quite constantly as part of LLM's working principles.

I mean, it's been more than a year since I decided to stop using Youtube altogether (I don't even use alternative front-ends and I actively ignore whatever Youtube link I may see/receive), so I can't say about Youtube's actual translation (I have only heard about it).

However, Google Translator is something I use quite frequently (even for English language, which I'm able to speak/read quite fluently; I use Google Translator in an attempt to verify and fix my grammar, given I'm Brazilian with a Portuguese-speaking inner monologue so I tend to formulate texts with a romance-y structure which may end up grammatically incorrect).

And it's been a while since I begin noticing how it changed (it became worse). Translations have been lacking clauses and sentences present in the input text; it's been struggling with parenthetical asides and texts between quotation marks; it's been struggling with romanization (i.e. transforming the approximate spelling for non-Latin languages such as Arabic, Hebrew and Chinese (three languages of which I sometimes tinker with for religious/esoteric purposes) into the equivalent text using Latin letters), among other inaccuracies.

Then I saw some news outlet revealing how Google Translate have not just been using Gemini under the hood, but it's also been "answering" injected prompts, as if Google Translate were yet another Gemini front-end.

So it's been unreliable. Unfortunately, it's the only translating engine capable of dealing with Latin (which I also tinker with sometimes), and the only one (at least the only "free" one) able to deal with long (more than 3000 chars) texts.

Given how governments around the world (yeah, not just yours, USians) have been "pivoting to AI" (which, IMHO, is not a problem by itself; the problem is when it's pushed down everyone's throats and smuggled onto places where it shouldn't be, and/or used for nefarious purposes), it's not a matter of "if", it's a matter of "when": "What happens when AI gets it wrong".

Diplomatic translations are the least of concerns, when we remember how air traffic control, command and control for entire LEO satellite constellations (to keep them in orbit while trying to avoid a Kessler syndrome) and even the maintenance of deployed-and-ready nuclear arsenals have been pivoting to AI, too.

I can guarantee you: it'll not be some diplomatic mistranslation to blame, but the future will be M.A.D.! 😬

'This is not a show': Macron criticizes Trump for Iran war and NATO comments