There was a time when Google Translate was the only game in town but now every single time, I find DeepL to be far superior.
This can happen to OpenAI? If not, why not? Is some of the research not open?
Also, even if DeepL is subjectively better, how did they or someone can do that? I mean some key papers+data+cloud budget to train are the three main ingredients to replicate such a translation model? If yes, that's also applicable to GPT-4?
To me, deepL was really amazing back in 2020 and earlier. But ever since big AI releases like ChatGPT, I can't help and feel they're kinda stagnant.
Their model still has the same incorrect repetition issues it's had for years, and their lack of any kind of transparency behind the scenes really makes it feel like it still _is_ the exact same model they've served for years.
Quickly checking their twitter does not seem to reveal many improvements either.
Of course I get that model training can be rather expensive, but I'd really have thought that in the field of AI translations things would be evolving more, especially for a big-ish player like deepL.
If anyone has insights on what they've been up to I'd be really interested to know.
Same, feels like they created a model once and left it to rot. What baffles me is how google translate is still behind them despite spearheading the scene a few years ago.
If you're asking "Could another company make a much better version of their products?" then the answer is a fairly clear yes, it could happen to OpenAI. They have an unknown number of proprietary advancements that give their products the edge (IMO), but it's an incredibly fast moving space with more and more competitors, from fast moving startups and deep-pocketed megacorps.
The impression I get is that if they rested on their laurels for a fairly short time they would be eclipsed.
ChatGPT is pretty amazing for translations as well. Even better than DeepL, since I can not only ask it to translate but to also change the tone and stuff
This can happen to OpenAI? If not, why not? Is some of the research not open?
Also, even if DeepL is subjectively better, how did they or someone can do that? I mean some key papers+data+cloud budget to train are the three main ingredients to replicate such a translation model? If yes, that's also applicable to GPT-4?
EDIT: Typo + clarifying question