They should combine it with Whisper so I can talk to some AI in a random language and just talk about nonsense. Also awesome training data since my pronunciation would suck? I am a bit too embarrassed to talk in a new language but staying with an AI would be ok.
Duolingo Max, a learning experience powered by GPT-4
41–50 of 250 posts
Re: Duolingo Max, a learning experience powered by GPT-4
#42Earlier quoted context omitted.
My girlfriend is Spanish. I don’t think her family would be impressed if I were to pronounce “gracias” as “grassy ass” xD
What's wrong with it? I'm a native speaker of neither, but they do sound similar (of course, gracias being pronounced South American style, not Spanish style). Except the fact that Spanish and English "r" differ, but I think it's reasonable for English speakers to know that, once they learn to pronounce Spanish "r".
English is a stress-timed language whereas Spanish is a syllable-timed language [1] - there's just no way to square that circle pronunciation wise.
Re: Duolingo Max, a learning experience powered by GPT-4
#43Earlier quoted context omitted.
Like many current LLM results, they're impressive considering the SOTA of ML until recently, but not very good in comparison to a good human baseline.
What example English words would you give as a guide for pronouncing the words?
Good question, it's not that easy to find existing words that match that well. If you asked me to do this I'd probably try making up fake words that in my mind, at least would sound closer (and that is also non-trivial given the weirdness of English pronunciation).
Also, I'm no linguist, my only credentials here are being bilingual, albeit with an accent (lived in the US from 0 to 8 years old and the past ten years, and a south-american country from 8 up to my 20s). So I lack the terminology to describe what things sound like.
That said - Cola, I think is fine. The "o" in toes is pretty different from the one in adios. I'd say that maybe it rhymes with "adipose" or "lactose", except the accent goes on the last syllable. Gracias, isn't too bad, but maybe I'd say "fascias" (admittedly a relatively obscure word). Manana is obviously wrong - here I'd just make up something like "man-nyah-nah" (accent on the nyah part). For uno, I guess I'd just say it sounds like "boo-no" without the b. Tres is close enough (the T and the D are different, but it still rhymes to my ear).
Re: Duolingo Max, a learning experience powered by GPT-4
#44This is interesting because my impression is that LLMs perform poorly on non-english languages[1]. This is likely 'just' a corpus size problem - but it's not like that is going to be fixed any time soon. That said, at least in language instruction there's a lot of meta material that will be helpful. I wonder how Duolingo is detecting when GPT-4 is hallucinating. [1] https://twitter.com/asmelashteka/status/16309662332…
They don't perform poorly (depends on how much corpus training the language received). https://arxiv.org/abs/2108.13349. The important thing is you don't need an equivalent number of tokens for an equivalent performance. So even though GPT-3 only had 1.8 % french by word count, it can speak fluent french. There's positive transfer between languages.
Also GLM-130B trained on equal English/Chinese (200b tokens each. 400b total) speaks both just fine.
Re: Duolingo Max, a learning experience powered by GPT-4
#45Just take a look at how poorly the bot does for Spanish: https://news.ycombinator.com/item?id=35155577
What is wrong with the example? Although I am not a Spanish speaker, naively it looks fine. The pronunciations written in English are very impressive and do indeed sound somewhat like the corresponding Spanish words (albeit of course incorrect! But most people don't know IPA).
No, it's absolutely terrible on every single one. For some reason it's telling you how to pronouce them with an English accent, or rather how an English-speaking person would pronounce it if they didn't speak Spanish at all.
> Adiós: "Adiós, my toes" - rhymes with "toes"
In this example "toss" would be much closer than "toes".
But yeah, this is awful. I don't know how this happened.
Re: Duolingo Max, a learning experience powered by GPT-4
#46Earlier quoted context omitted.
Irresponsible to give this to learners
My girlfriend is Spanish. I don’t think her family would be impressed if I were to pronounce “gracias” as “grassy ass” xD
Well, if you read it with a Northern English accent like a Mancunian accent it's pretty close!
Re: Duolingo Max, a learning experience powered by GPT-4
#47Earlier quoted context omitted.
My girlfriend is Spanish. I don’t think her family would be impressed if I were to pronounce “gracias” as “grassy ass” xD
It does kind of sound the same tho? Grassy-ass. Grassiyass. Grassyaas. Gracias.
If you drive from Virginia to Boston, you'll probably hear 5 different pronunciations of "Grassy ass". I don't think any of them sound like how I learned to say "thank you" in high school Spanish class.
Re: Duolingo Max, a learning experience powered by GPT-4
#48This looks great! I would like to know how much this costs though.
Re: Duolingo Max, a learning experience powered by GPT-4
#49Earlier quoted context omitted.
I also use it daily but I'm a bit worried I won't concentrate as much if I have unlimited hearts, have you noticed any decline in recall? But yes, the ads are super annoying... especially the same one to upgrade Duolingo puts in every time!
Just starting my second paid year - I didn't put up with the adverts for long. Works far better for me without the hearts. Less time staring at sentences trying to make them perfect - more quick failing, learning why, iterating and drilling it into my head. Also much more likely to spend longer in the app, maybe a couple of times a week.
After a few months of Duolingo you should move onto comprehensible input for learning. Your Duolingo subscription is holding you back
Re: Duolingo Max, a learning experience powered by GPT-4
#50Earlier quoted context omitted.
Like many current LLM results, they're impressive considering the SOTA of ML until recently, but not very good in comparison to a good human baseline.
What example English words would you give as a guide for pronouncing the words?
That is the way language learning books were written before the days they could provide audio recordings. Now with modern technology it is much better to learn the sound system of your target language then try to find comparisons with your native language.