Kagi Translate lite/eco

Thibaultmol

While I love Kagi Translate and how it does indeed seem to be DeepL or above (depending on language) level.
I don't always need that. Heck, i reckon to say that I usually don't need that.

I love using Kagi Translate when I need to translate for something that I need to send to someone.

What I don't want to use kagi translate for is translating things to one of my own languages. I use DeepL or Google Translate when translating a webpage or a piece of text if I need to understand it or get context for it. I haven't felt the need to get more accurate translation than those for that application.

So this is why from an energy usage standpoint: i'm not going to use kagi translate for these types of applications. i want to use something lower energy usage.

There are some open source projects for translations. Are they lower quality? 100% def.
But do i need highly accurate translations if i just need to get a quick understanding of a page or something. no.

It would be nice to have an option with kagi translate to use a non-llm system as well.

@marc_vb suggested https://kagifeedback.org/d/5305-kagi-translate-feedback/39
something like https://lingvanex.com which is what Vivaldi uses (I haven't found that they indeed host their own server, but they could).

or some other (maybe FOSS) project could be used.

But for webpage translation and other "I just need to get a basic understanding of a piece of text in another language" I won't be using kagi translate because of the energy usage it causes. I'll be using DeepL or Google Translate for that.
But having an 'eco' mode in Kagi Translate would be nice because then it's all in one place

marc_vb

Thibaultmol (I haven't found that they indeed host their own server, but they could

https://vivaldi.com/features/translate/

The translation engine is securely hosted on Vivaldi servers in Iceland. This means there are no third-party servers involved. And it means nobody can access your private translation activity, even us.

NB: Good to know that Vivaldi refuses to use AI.

nichu42

Thibaultmol Would that be different than just using https://translate.adminforge.de/ or a similar page?

Thibaultmol

marc_vb * I'm curious what Vivaldi says about their translation system because that does use AI. because of course it does. Any modern translation including Gtranslate and deepl use machine learning or transformer networks to some degree.

What Vivaldi mainly means in the blog post is that they don't use LLM's. Which is fair.

But I personally haven't looked into lingvanex that much, but it seems like might be using similar components to LLM's:
https://lingvanex.com/en/blog/what-are-large-language-models/

Lingvanex uses the OpenNMT-tf framework for its translation models, which are based on the classic Transformer architecture (encoder + decoder). This approach ensures high-quality translations and optimises the training of language models.

Meanwhile libretranslate uses machine learning but not LLM's as far as i understand: https://libretranslate.com/

marc_vb

Thibaultmol because that does use AI

In these days, for commercial / investor pleasing reasons, everything is called AI. Local ML is called AI. Algorithms are called AI. If-then-else is called AI. Etc.

I don't know for sure if Vivaldi is using Lingvanex's AI capabilities; maybe that's just an option. Knowing Vivaldi a bit, I don't think they have a heavy server infrastructure for it.

https://lingvanex.com/en/blog/cases/business-case-20/

Bottom line: As long as "AI" isn't using LLM's and/or is using a lot of resources, then I'm fine with "AI".

carl

It's not good business sense to associate your company and brand with low quality products, especially if you are funded by user subscriptions. It's also a bad experience for the users, who have no way of knowing what quality to expect. So I don't like this idea. If you are that adamant about saving a tiny sliver of electricity, I think an old fashioned paper dictionary could be the best option.

Thibaultmol

carl I don't appreciate that last sentence. You could have commented without that....

People understand that eco is normally worse than the normal thing. I'm just giving ideas

Also: kagi translate is free. You don't need an account. I don't agree with your first argument

marc_vb

carl a tiny sliver of electricity

Yeah right.

By 2026, the agency projects that data centres’ energy consumption will have increased by between 35% and 128% (mostly due to AI)

carl

marc_vb One single persons use is a tiny sliver. Utility computing (such as translating) is extremely effective in comparison with other normal household uses of electricity by an individual.

Thibaultmol

Sorry for the tone, but I really think that what you're suggesting is akin to straining gnats and swallowing camels. And I think that Kagi should avoid connecting their brand with low quality offerings. Even in a free product, which can be a gateway for paying customers.

marc_vb

carl Utility computing (such as translating) is extremely effective

Computing can be done with low energy consumption; sure. But AI, especially with LLM's, involves heavy use of resources like water and electricity. Simple facts, documented by OpenAI, Microsoft and Google, to name a few.

Thibaultmol

nichu42 it wouldn't I suppose, but that's assuming that people find their way to those sites.
Plus once Kagi translate is in the extension and such, it'd be nice to have an eco toggle. while libretranslate would need some other extension, if these demo instances even have api access allowed to begin with

gladiator2339

carl

marc_vb All computing is low energy usage, compared to what we'd have to do without computers. It's not as easy as measure as heating or transport, but information technology has immense utility. Especially translation.

SonephetR2

carl The Jevons paradox teaches us that sometimes "when technological progress increases the efficiency with which a resource is used", total "resource use is increased, rather than reduced" due to increases in demand. Casual translation seems to fit the bill.

Firefox Translation does a great job for basic, on-the-go translation needs for the few languages it handles for instance, and is entirely local. 3 MB covers 10–17 languages.

marc_vb

carl All computing is low energy usage

Sigh. You really don't want to understand? You're not aware (or don't want to be) of the huge increase in energy usage due to AI / LLM's?

carl

marc_vb It could also be you who don't want to understand. It's not Kagi's responsibility to stop people from using their product in a wasteful manner. That's the concern and prerogative of the customer. If you want to compare energy usage of LLMs, you have to compare to the energy usage of having a real human translator. I used to hire and pay people for translations. That energy cost is probably 100-1000 times higher than a LLM.

It you're selling a premium product like Kagi, it is bad business strategy to associate your brand with low quality products. @SonephetR2 suggested Firefox Translation, which I think is a great solution for OP.

marc_vb

carl That energy cost is probably 100-1000 times higher than a LLM.

Aaargh, forget it. This is of nu use. If you keep making claims that are factual false, then this discussion is pointless.

You may react to this if you want, but I won't.