Comment by fsuts
4 days ago
But fair play to OP, you'd have to bounce it through at least Google Translate to make it legible to the predominant English-speaking HM crowd: https://x-com.translate.goog/akiraa_tanaka/status/2080114490...
People still use google translate? Most browsers now have translate built in, as do llm
I imagine grok on x will do this too
X’s auto translate of tweets has been really awesome actually.
I wonder why its so much better than localised titles and subtitles on youtube.
LLMs can mentally match and replace whole sentences with culturally similar concepts rather than having to go word-by-word. That tend to yield more natural sounding, not necessarily accurate, translations than literal ones. That's great if you don't need accurate details and/or consistent inverse translations.
Classical machine translations can't take "EUTs sets up a bearer with eNB" and produce "The device (like a phone) connects to the cell tower."[1], it's a fragment of confection to an LLM, and that's useful when/where it is.
1: I just literally asked Gemini Flash to do it
Firefox didn't offer to translate this, so initially I wasn't aware that it could. But I could get it to translate via the menu. And Google Translate is still a bit better, so I do regularly fall back to that.
Unless something changed recently (and things change all the time where LLMs are involved), deepl.com does the best translations. I have a sister who translates professionally; she said it's the only one where she can use the result and make some improvements instead of throwing it away and starting over.
Deepl's professional advantage is not in the translation itself (even though it has always been a notch above the rest), but in the UI. You can literally click any word on the translated text and (a) see a list of alternatives, and (b) pick one and then the following text adapts to your choice.
In technical terms it lets you stop the transformer at any token, sample alternatives from the probability distribution at that point, pick your desired next token and resume the generation.
This UI lets an expert guide and control the output of the tool, instead of having to fight it. I wish there was a frontend for large GPTs that worked like this instead of the ubiquitous-yet-useless chatbot.
2 replies →