Can AI Correct Your Japanese Mistakes?
Вставка
- Опубліковано 28 чер 2023
- Learn Japanese with Yuta: bit.ly/3raGjvw
Support me on Patreon: goo.gl/aiWNd5
Twitter: / thatyuta
Instagram: / thatyuta
Facebook: bit.ly/381qpHS
Blog: www.yutaaoki.com/blog/ - Розваги
Learn Japanese with Yuta: bit.ly/3raGjvw
Sounds like these AI need to join Yuta's mailing list, so they can learn the kind of Japanese that real life Japanese people actually speak!
that was really funny 😂
GPT has plugins, so someone should make a plugin that allows GPT to join Yuta's mailing list
👏😆👏
6:06 I have an anecdote about that.
I had an encounter with a dude on a discord server saying to everyone that he was fluent in japanese and was living in Japan for 5 years. I was interested in practicing japanese so I started to chat with him a bit, but his sentences felt weird and overly polite (the first sentence he said was : 私は元気で、あなたは？) that's when I suspected he was using an AI. I tricked him by saying I couldn't understand polite japanese and asked him to speak casually, which he was unable to do, then I wrote very very unnatural sentences and asked him if he could spot my mistakes because I wanted to improve my skills, he pretended that everything was good. I called him out and he left the server.
Lmao I love calling out people like that, it's so fun
Some things never change. I've seen this happen so often in Discord and when I said on my dating profile I was studying Japanese so many dudes tried pretending to be fluent with Google translate
@@sucyshi what is their issue? Are they pathological liars? Really dumb.
Sorry for the dumb question (trying to learn Japanese, basically just started), but is that first sentence wrong because it's very English-like in the way it's structured?
@@sune9578 It’s because it’s unnatural to ask people if they’re 元気 in Japanese unless they’re sick. It’s not really something that Japanese people say, unlike what the textbooks want you to believe. Maybe it’s because having the “how are you” from English will make Japanese look less scary, but it’s more like “are you well?” for a sick person (お元気ですか？ is a common thing that beginners say.) Also, starting a sentence like that with 私は is unnatural. In addition to that, that usage of で is extremely weird. While yes, you could translate it as “I’m well, and you are?”, it’s not something anyone who is actually fluent in Japanese & living in Japan for 5 years would say. Definitely plausible that an AI would write a sentence like that.
Yuta trying to go Super Saiyan with that hair. Lol.
Well with all the frustration Bard was giving him Yuta is close to go super Saiyan 😆😆😆
It looks good though.
I was actually thinking Jimmy Neutron
It gives me MacDonald's ice cream vibes
Something that I noticed throughout the video: Ai neither asks you what you really want to say in your native language before correcting the given sentence (so it would compare structures and words), nor asks you what your native language is before it gives you an answer.
That's one of the things I've found gives better results with these AIs -- the more context you can give upfront, the better.
@@guspollywith ai art can take 50 prompts to get what you want
I miss the little jingle that used to play at the start of your videos. It gives me the same joy of hearing about a Japanese man named Yuta who can teach me how real Japanese people speak.
*YUTA-chan* promoting his mail list is always so cute.
Makes me smile.
It's so annoying lol
I can't believe Yuta was able to place the Monogatari reference in this video 😆😆😆 soooo amazing
For those learning Japanese: I have been using GPT and it's an amazing tool, if used correctly.
Let me explain how GPT works, in simple words, so you can make the most of it:
GPT isn't actually intelligent, it doesn't understand your sentence. What it does is completing your sentence with the most probable words that would follow it. So, when you write a sentence or a question, GPT finds the most probable word after that, then the most probable word after that one, and so on, until it has produced one or a couple of sentences more.
So, you need to find the sentences that work best with GPT, for your needs. You should ask questions in a way that is common, in a way many people usually ask them, so GPT will find a good answer for those with high probability.
If you ask weird questions, uncommon questions, or sentences that rarely people say, GPT will produce a weird or random answer.
It's like, if I told you "dog what red", what would you say? The continuation of that sentence could be anything, like:
* green blue
* maybe you mean what dog is red?
* what blue
* sorry, I don't get it
Alright mate can you explain how you use I've not used chat gpt
@@ashmorris4067 first just try simple questions like "how to say good in Japanese", or copy some Japanese sentence and ask "Can you explain this sentence: いいですよ".
but you can ask more complex questions like "what is the difference between these sentences: ...", or "when should I use が or は with the subject of a sentence".
I ABSOLUTELY HATE the search--I ALWAYS have to read EVERYTHING that has ABSOLUTELY NOTIHING to do with what I'm searing for. Thanks for demonstrating; I feel your pain🤗
I missed you yuta!!!! Last time I learn japanese with you was in the quarintine time. And now i’m continue learning
Your hair looks great Yuta! And this was a very interesting video, thank you.
I am no where near approaching Japanese, but as a language learner of Spanish French and Russian, Japanese is becoming a language I am thinking of tackling but it seems very difficult to learn outside of Japan. No apps have the full comprehension you need, and unless you can use authentic Japanese resources I would imagine ti could be hard. Thank you for the effort you put in to teach us natural and realistic usage of the language. (Even though I’m not learning yet)
To be honest there are plenty of resources to learn Japanese. I liked to use Tae Kim's grammar guide coupled with Wasabi (Learn Japanese Online) and Maggie Sensei.
The key is to learn from different sources and get different explanations for different things. ChatGPT is honestly a good learning partner for when you struggle with some concepts (even though you should be careful about its answers).
Given the abject failure of dedicated translation apps to deal with Japanese, I really doubt generalized ai is gonna cut it. However, given the self correcting algorythms used in ai, it is quite likely that an ai program specifically fed a diet if correct japanese would probably master most of the language pretty quickly. Whether it will be able to read the context, let alone read the atmospere, is another matter. Get ready for aspberger's ai.
"Get ready for aspergers ai."
OMG that's hilarious!
I have checked your attempts now on both chat GPT (free version 3.5) and bard. both did a good job. So AI is getting better quickly everytime you go back and check it
Unfortunately, I don't think ChatGPT can fix my birth. lol
mood
Bruh
最近、大学で経済経営学を学び始めたんだ。更に、大学生チームに入って、大学のイベント開催するスタッフにもなった。特に、慣れてないトピックを話すと自分の日本語が若干不自然に聞こえるようになってしまうことを最近気づいたわ。
Yuta will teach the real life Japanese that real life Japanese people actually speak? I just want to confirm this because I only heard it a couple times in the video.
LLMs in general are pretty inconsistent. You could try to regenerate the answer a few times.
The power of LLMs in my opinions lies in fast iterations through feedback. I'd like to see what happens if you ask it for improvements if needed. For example, say:
"I like that you corrected the grammatical mistakes, but the sentence still sounds unnatural. Can you make it sound more natural?"
OR
"Ilike that you corrected the grammatical mistakes, but you converted the phrase from neutral Japanese to formal Japanese, and I'd like the phrase to be in neutral Japanese. Can you change that?"
You can just try it out yourself. I do use chatgpt 4 to help me with natural sounding Japanese and help my confidence for stuff like a business bio and planning out what to say in a phone call to be as clear as possible, but to do that I have to coax it to stop using so much textbook Japanese.
That said if I wasn't around N3 using it like this would be pointless as I wouldn't be able to tell when it's being too textbooky, formal, or casual. And I'm sure I'm not catching a lot of its mistakes but I'm hoping it's still better than me by myself
@@sucyshi I don't know any Japanese unfortunately, so I wouldn't be able to tell what was wrong, or even come up with feedbacks for it. I use it for coding every now and then though, and that process is how I do it. Normally I ask for something, it outputs garbage code, I tell it what is wrong and what to fix and it does that, rince and repeat until I have something useable.
@ThatJapaneseManYuta Can you take a look at Satori reader and check how natural the stories and conversations there are?
This is a small aside because everything here definitely gives the viewer the correct idea about which LLMs are able to handle this task, however you really should be starting a fresh prompt for each question because they get more confused the longer you keep the conversation going.
The sad thing is AI still does a better job than local American translators.
Turns out that chatgpt can't teach the real life japanese that japanese people actually speak!
Yuta, what about the idol phenomenon in Japan?
That kind of relationship that science calls "para social".
Recently Saya Hiyama (weather girl) had something similar when she revealed that she was dating. I myself was on this "wave" for a couple of months.
Oh man you got me .
YUTA-san! Have you ever heard about the 'Isle of Dogs' movie by Wes Anderson?
Why replace madoka with chainsaw man?
yuta mate you've plugged your Japanese lessons at 20 seconds in are you trying to get some world record 😂
Sup
How to tell 'how can I help you?' in Japanese, when someone asks for ur help...
If you want to hear a more casual grammer suggestions, simply instruct the AI platform to use it. I find that comunicating with these platform is an iterative process. You don't always get your desired answer on the first try.
is there a japanese grammar/spellcheck checker like for english instead?
You can use bunpo-check. It's not perfect but not so bad
日本に行きたいだよ？
You also don't use desu masu when you're just thinking it to yourself.
Yuta, is ramen good for a diet?
Not if you're trying to use weight. Ramen has a lot of calories and salt so it's not usually very healthy to eat it a lot.
@@coolbrotherf127 Really? How about sushi?
@@name3583 most kinds of sushi is good. Especially if it's just the fish with rice. Sushi rolls that have lots of other stuff like mayonnaise or cream cheese can be unhealthy though if you eat it too much.
Isn't instant ramen the college student diet
Generally the most popular Japanese foods abroad are their fast food. Ramen is noodles (carbs) and bone broth soup (hella fat). There are lower calorie ramen broths but it's still mostly just carbs.
Healthy foods will focus on proteins and/or veggies with few added carbs, fats, etc. So sashimi is definitely good (do your research on stuff like mercury in fish though), and Japanese sushi (not western sushi) with lower rice to fish ratios are good. Carb intake is still important, but focusing on preferably whole grains and eating it alongside proteins and veggies is best to reduce hunger.
Miso soup is also good. Japanese pickles are traditionally eaten with most meals and are great as a veggie source and for gut health. Tea with your meal can also help you feel more full, mostly due to it being a warm drink
dude is typing in light speed
Try TELLING AI to explain in English rather than asking if it can.
Don't forget IA are tool, and to have a great output you should do a good input. Using ia correctly aka creating prompt is a work, please next time try to give more context to the IA, and i'm sure their response would be far far better.
AI is very strong at chess game. I never defeated AI in chess.
You don’t need AI for that, computers have been better than all humans at chess for 20+ years now
@@somebodyuknow2507 What's the difference between AI chess and computer chess?
@@name3583 Traditional chess engines that have been better than top grandmasters for years are, though complex, still a typical computer program where the program takes a position as input, executes an algorithm that the programmer made, and outputs its move and an evaluation of who is winning.
Neural networks, which is what people typically mean when they say AI these days, are different. They also take a position as input, run an algorithm, and output a move/evaluation, but the difference is that nobody knows what that algorithm is. They make their own algorithm during training by modifying their internal structure. By the end, though we can see each neuron and its connections, understanding the structure of the whole thing is impossible, similar to human brains.
The strongest current engine is Stockfish, which is a traditional engine. Second place is Leela, a deep neural network (AI) engine.
TLDR: you don't need a deep neural network AI to make the strongest chess engines, a traditional computer program works just as well.
@@somebodyuknow2507 I don't understand a lot about chess engines, but don't the latest versions of Stockfish use NNUE?
🌹
AI has its problems.
It's a hard task for the machine learning models.
Maybe try something they actually should be strong at?
Perhaups if I'd write a paragraph in English and have it translate it to Japanese with added instructions on whom (friends or strangers or boss) to adress?
Or even better, have chatgpt write a short paragraph about any topic on its own from scratch with again the added instructions of whom to adress?
Having chatgpt translate and explain is hard because of mathematical and probable distances within the trained model.
See, when they feed the model with texts in various languages, you can think of the learned things as "bubbles".
If they hadn't fed the model with any translated texts at all, and every text was unique, the trained mathematical model would not be able to diliver any translations at all, because the "bubbles" never overlap or touch each other.
You can think of the translated texts to act like a bridge between the "bubbles". Or in mathematical terms the translated texts move the sets closer to each other raising the probability that text A maches text B, hence resulting in a probable translation.
But! If you bypass this little problem by sticking in the "bubble", you should get pretty good texts.
But why? Well, what do I care about the text book "Peter, Paul and Mary watched a football game." when I could not care less about football?
Some people love the tedious repetitions and even more tedious example sentences. But I'd propose not a single living babies soul on earth nor history has ever learned their mothers tongue by grinding through irrelevant text snippets.
We all learned because the words we learned had meaning to us at that point in time.
Hence, having chatgpt spit out a paragraph about that dude making sweet loving quality time with that beautiful redheadded woman will (so I propose) stick much wicked more than any dumb text about any shopping situation.
Is the latter unimportant? No. But it'll come automatically on its own and in time.
Now, when you have the Japanese paragraph you can easily rip the sentences apart, throw them into any text to speach app and hear what it sounds like.
And in terms of grammar, do 6 year young children know much about grammar? They do not. Are babies tormented with grammar?
Why then should we not have fun with languages and do the grammar later?
And if you object to all this then you can easily save teachers valuable times by having them cross check the chatgpt text, then learn to speak it, and afterwards have the teachers explain the grammar if you desire. Maybe by that teachers get to keep their sanity by not grinding through basics 90% of their time.
I'm an old school guy who learns things by figuring them out on my own with some guidance, artificial intelligence is not my cup of tea and never will be.
figuring things out on your own with some guidance seems like a perfect use for AI tho (specifically, these large language models). it can give you very good guidance in the basics of most things if you prompt it do so. just saying
@@varencilator It really can't.
First
first
AIは下手
下手くそ！
Copy and paste...
It can barely speak english normally
This channel is slowly dying
Not more AI shit...