App resource translator

[veler]

What feature or new tool do you think should be added to DevToys?

This is an idea of extension for DevToys 2.0.

As an independent developer, I might need to translate my app in many languages that I don't speak. Many tools can help for this, such as Crowdin, thanks to community contributions. However, it would be nice if we, developers, could provide a first, approximative translation that people fluent in that language can just review later. This can help getting an app translated faster.

Why do you think this is needed?

It would help independent developers to get their app translated faster. For instance, some DevToys 1.0 languages are only translated at 10-15% due to lack of contribution. Having a pretty tool automating the translation could help a young project to reach more people.

Solution/Idea

To solve this issue, we could imagine an extension for DevToys 2.0 that would use Chat GPT to translate automatically some RESX, RESW, RC and other resources files automatically. Chat GPT already proven itself to be quite efficient at translating large text pretty accurately, better than Google Translate, Bing Translator or DeepL.

Comments

No response

[RonStrauss]

This sounds like a promising idea. I think it could be done by utilizing one of the many trained LLMs up on hugging-face. Or maybe use ollama.

ChatGPT would work well with an agent, but since it costs money, I don't see many developers jumping on it.

The prompt needs some work, but it might be worthwhile, using ollama:

[veler]

Thanks @RonStrauss ! I already have an issue that tracks the use case of a local LLM. https://github.com/veler/DevToys/issues/825

There's one thing that ultimately worries me about it: how slow it is despite powerful CPU. Yes, GPU improves it, but not many professional developers have high end GPU on their work machine. I'm concerned the intense work on the CPU may "fry" some of them. While we reject any liability for damaging the customer's computer, we don't want to get the reputation of damaging them 😅

Ultimately, perhaps we can offer 2 versions:

1. A slow, heavy and less accurate version but that respects privacy since it works offline (which is the main argument for using DevToys initially).
2. A faster, lightweight but paid version that use Chat GPT (and so internet).

[RonStrauss]

Thanks for linking that @veler 😀

I guess I'm a bit out of the loop, you guys are way ahead on all the implementations for it. Frying CPUs does sound like a bad idea lol