It does contain Finnish to English and vice versa
https://mobiili.fi/2022/07/06/meta-julkaisi-avoimena-lahdekoodina-200-kielta-kaantavan-tekoalymallin/
They have given the source code for their AI as opened sores so that everyone can have it.
Its able to translate 200 languages from one language to another. Any of these languages can be either a source or a destination.
However it does back up into English in all cases..
Finnish to Spanish? It goes to Finnish-English and then English to Spanish. Same for all other languages too. Speed is nevertheless instantaneous.
It is a text to text AI and not a speech interpretation program.
Seems like a cool idea
>However it does back up into English in all cases
Aaaaand now it doesn't anymore
What are the hardware reqs for inference? How much video memory?
Mene takas ylikselle
haista paska
They commented they were working on a direct translation model (no english in the middle) at one of their press conferences. I would guess they are releasing this because they have a better solution in the lab.
Direct translation makes more sense since words in different languages never overlap 1:1, they always exist in some foggy latent space with vague meaning
Sex
Coffee may be good for you in some doses, but you can't absorb b vitamins properly. Also, it can give you anxiety. I prefer green tea because it has L-theanine.
Wrong thread, anon
newgay
i hate ML
with conventional problem solving you develop clever methods and deep understanding of a subject matter to fight for the smallest results
ML on the other hand can achieve impressive results in some cases with just brute force and not actually advancing the field it's utilized in
it just feels so wrong
heh first experience with the bitter lesson?
the greatest observation ever made is simply moar cores and moar ram can solve any sufficiently complex problem better than you can solve it
This is why I don't get why people are trying to push for more ML in scientific research. That's retarded, the whole point of research is to understand things, not to produce a black box model with a bazillion parameters.
because they want to compound understanding by producing a black box model with a bazillion parameters that is smarter than a researcher
How does this compound understanding? This is so vague, sounds like you just fell for the hype.
AGI is by nature an alluring hypothetical
welcome to the real world
lmaoing @ the seatbelt clip
Imagine the same clip but on her panties
>snowmoron babble speakers mad that a project prioritizes the only language that matters
classic
hope solo. she has nudes.
>opened sores
she fucks black men doesn't she...
>black
>men
morons aren't human dummy
Do you know what separates humans from animals?
the mediterranean sea
Finally somebody gets it!
>italians
>human
italians are human, but they are not white
where's her boobs?
cool beans
it used to be that I can apply the 'telephone test', ie translate X to Y then back to X multiple times to see if there's any degradation. But I think most translation services caches all results now to make it bijective by definition so the test doesn't even work
Name?
Maria Kaczyńska