The only use for machine learning I have is translation software, which was getting pretty good already before the LLM hype. Assuming I’d get that back at the pre-hype levels I’d have zero impact.
I only use it for checking how it would propose to formulate a sentence in French, but then I write my own anyway. Since it would be weird to pretend that was my level of skill towards the customer. Or sometimes I translate back my wobbly French to see if I went really wrong or ran into some false friends (terms that sound similar among the languages but have different meaning).
It’s like consulting the Chess Engine for the engine move to me.
If this software too vanished, then I’d just have to use the online dictionaries even more. And be even slower, but who cares this is only once every few weeks, anyway. Most French speaking Swiss are fine with English in text form too.
We have gone from reasonably-okay translation programs to “intelligent” systems that are flat-out incorrect. And somehow people believe that the incorrect ones are better.
I look forward to this particular hype train running off the rails sooner rather than later.
Yes for Google Translate at least. It’s still possible to inject prompts in your “translation”, too. I’ve noticed a considerable drop in accuracy translating to English, too… The English is more natural but “forgets” to include some information in the original text. This stuffs up my old workflow because now I can’t rely on Google translate at least translating all the nouns (e.g. it will drop some names in a list of people)
The only use for machine learning I have is translation software, which was getting pretty good already before the LLM hype. Assuming I’d get that back at the pre-hype levels I’d have zero impact.
I only use it for checking how it would propose to formulate a sentence in French, but then I write my own anyway. Since it would be weird to pretend that was my level of skill towards the customer. Or sometimes I translate back my wobbly French to see if I went really wrong or ran into some false friends (terms that sound similar among the languages but have different meaning).
It’s like consulting the Chess Engine for the engine move to me.
If this software too vanished, then I’d just have to use the online dictionaries even more. And be even slower, but who cares this is only once every few weeks, anyway. Most French speaking Swiss are fine with English in text form too.
As a translator who changed careers because of CAT tools: fuck that shit. Translation agencies have used them to strangle the market. End of my rant.
Maybe don’t ask LLMs for translation help. Apparently because many of them are programmed in English, they still follow English structures when writing in other languages, resulting in products that are grammatically incorrect.
We have gone from reasonably-okay translation programs to “intelligent” systems that are flat-out incorrect. And somehow people believe that the incorrect ones are better.
I look forward to this particular hype train running off the rails sooner rather than later.
Well I haven’t changed the tools I use (DeepL and Google Translate), but I assume they built in LLMs in the back end.
Yes for Google Translate at least. It’s still possible to inject prompts in your “translation”, too. I’ve noticed a considerable drop in accuracy translating to English, too… The English is more natural but “forgets” to include some information in the original text. This stuffs up my old workflow because now I can’t rely on Google translate at least translating all the nouns (e.g. it will drop some names in a list of people)