[00:00] Hello my dear listeners this is English Canada the podcast about learning languages going through immigration and establishing yourself in a new place

[00:17] Hello everyone this is Anastasia Stoyanova with you guys welcome This is a new episode and today I'm gonna talk about something that's very important

[00:27] I think for everyone today and that's the use of AI and more precisely speaking the use of language large language models

[00:40] And recently I've been doing a bit of training so I've been trained in information technology and I'm learning about lots of things like computers you know programs and

[00:53] I'm trying to apply all of that knowledge and skills in my work in teaching and of course today's episode

[01:01] I want to dedicate to using LLMs large language models I'm pretty sure everyone has used ChatGPT My personal favorite is

[01:13] CLAUD actually I use CLAUD for all of the tasks that I do daily for my work for lesson preparation for studying learning new things for learning

[01:27] French as well because I'm sure by this time you have already know just you've heard of that that I've been learning

[01:35] French a lot If not please listen to or watch my previous episodes So in today's episode I want to talk about first of all just some tips which model to use which version to use paid or non paid how to work with the prompt also which skills it's better to practice with

[02:00] AI if you're learning a language Then I'll talk about my personal favorites that's CLAUD and Perplexity and also some other things like some tips and tricks what

[02:12] I do and some examples how I use the large language models to learn a language for example the French language

[02:21] But also you can use AI large language models in particular for so many things As long as you verify the output and you always do this what

[02:34] I call sanity check check the information check the output play around with the input So first things first personally

[02:42] I've been experimenting both with paid and non paid versions and I highly recommend using paid versions

[02:51] And if you're a student you can always claim that and you get a student discount it's very easy Even if you are self employed or even employed you have an employer sometimes your employer might give you benefits extra benefits to use

[03:06] AI as well and you can claim that as an expense also something that you are using for your self development let's say

[03:16] For me as a self employed English as a second language instructor obviously I use that for business purposes so

[03:24] I always claim that as my business expense So I do recommend using a paid version you pay but you get much better output

[03:32] If you just play around and you ask the same question you ask you know or you assign the same task to the paid version versus unpaid or free version of

[03:45] LLM that you're using it's gonna be way worse like the quality of the response when it comes to a free version

[03:54] You could check this out with different models and you know if you're a chat GPT fan right or I'm a fan of the

[04:04] Anthropic company and Claude in particular so I've played around with free Claude and paid Claude and it makes a big difference honestly

[04:13] So that's my first recommendation Then also how to work with a prompt First things first if you are learning

[04:20] English let's say or French try writing your prompt in English or in French of course When you write your prompt that's learning in itself because

[04:30] I noticed that for me as a teacher also as an English teacher I practice English just by typing very detailed prompts

[04:38] That's a lot of writing because you know you always want to specify the context which role you are assigning you know act as a what as a who as a

[04:50] French as a French as a second language instructor and explain to me as if I have no previous knowledge of what kind of topic what kind of context

[05:01] And then you write as many details how you want the output to look like Like you you're going to say for example write five examples and provide with the translation to your native language and so on and so forth

[05:17] So a lot of people are just sometimes are lazy when it comes to writing prompts But it's actually everything depends on your prompt

[05:27] If your prompt is very specific your output will be more successful more exhaustive like more full right

[05:37] So that's why there is even a separate like direction of working with LLMs That's prompt engineering

[05:46] So your prompt should be very context specific You have to exactly indicate what you want to see as your output

[05:54] And also you have to provide a lot with your input For example along with your prompt you might want to provide some

[06:03] PDFs or markdown files just to give the LLM more context And you might want to say like consider the

[06:13] PDF files I'm attaching and I want you to study these files scrutinize them and I don't know review them and prepare summary intent sentences with the most important things and provide me with some examples or whatnot yeah

[06:30] But of course it's a brief example If when I practice with my students for my lessons I promise you that

[06:38] I will give you some more examples how to practice the skills So during the lesson sometimes when we look at the text because

[06:46] I teach cell pit preparation so we look at a lot of texts a lot of reading And I always input the text and

[06:55] I ask a large language model that okay we have this question and the correct answer is this So you got to provide the correct answer otherwise you know of course a

[07:07] LLM might always hallucinate So you got to make sure that your teacher or yourself if you're learning any language by yourself you got to make sure you know what the answer what the correct answer is and try to kind of ask

[07:23] AI why is this correct answer Can you explain to me like I'm five right Or can you provide the justification from the text

[07:33] So for example you could use for reading right When you do the listening and you do the tasks so maybe you're listening to this podcast right

[07:42] And you want to just like ask about something some concept So you can always work with a transcript right

[07:52] A lot of audio files or listening exercises in the books on the platform for example if you're preparing for cell pit or

[08:00] IELTS every listening is a text basically And you are going to input that text into LLM and then ask it to break it down for you

[08:12] So you see both reading and listening can be practiced with an LLM In my case I use Claude Claude has many models

[08:20] They provide the models that are more ambitious that can work with files that could generate nice looking

[08:28] PDFs for you For example for vocabulary I always ask one of the Claude models to generate an HTML file for me

[08:38] And it's an interactive quiz that I use during my lesson And I also provide my student my client with such quizzes that are very personalized

[08:49] And in that way they can practice they can play around they can open this HTML file in the browser and it's all free and nice looking

[08:57] And they click around and there are some explanations that pop up So it's a nice way to test yourself how much vocabulary you remember if you're learning a language for example

[09:08] So you see you could practice reading listening vocabulary as a language system right You could practice grammar and give the context for example let's say explain this present perfect topic for me in the context of life experiences right

[09:26] You could even provide some snippets of the lessons like if let's say your teacher made some notes for you and you can discuss the notes

[09:37] And for this purpose there is even a more specific tool that I like using It's Notebook LLM that's by

[09:46] Google And that's also they have a free version and paid version And of course it's up to you to decide

[09:52] But my recommendation is always to invest a little bit of money and you get much better outputs much more quality

[09:59] Yeah and that's just great Also you got to understand that different LLMs have been trained to do different tasks differently

[10:09] So let's say you want to do research online and you need to I don't know I will give a life example like less technical thing

[10:18] But let's say you're choosing a movie right And you know that you have liked certain movies or series

[10:26] So for this purpose I highly recommend Perplexity if you're doing the research Or any cloud model but making sure that you're using research mode

[10:36] And so by providing such examples and giving more context maybe like you know why you liked this specific movie

[10:46] And then in that way the model can do some research for you and make some suggestions So that's why like be careful because every model has been trained on certain data

[11:01] And so you want to use that specific model for a specific purpose that you have in mind Because AI is more like an umbrella term right

[11:13] But LLMs are really great when it comes to writing your code for example if you want to practice programming or teach you how to write code right

[11:22] Or at least the basics But you always always got to make sure to double check the output The same goes with language learning

[11:31] When they generate the output sometimes what I notice is that even the paid versions might make some mistakes due to some statistical analysis that they are using

[11:42] Like just a recent example I had with my husband's brother We were just discussing something online and

[11:49] I was giving some tips to him And I provided the list of the words and I said like okay look them up ask

[11:57] LLM to generate examples for you And so he did And I think he was using like a free ChatGPT and it actually made a mistake

[12:07] It hallucinated and it corrected the phrase spick and span because it's spelled as spick S P I C K So

[12:19] ChatGPT said like oh it's not that It's speak but it should have been spick and span Even the sound is it's shorter sound

[12:29] It's not speak like a long tense vowel It's laxed vowel It's short vowel Spick spick So it's a different expression but

[12:38] ChatGPT corrected him and corrected my husband's brother and said hey you must have made a typo it said

[12:48] But there is such an expression So you see you got to be careful And my point you got to verify the output

[12:57] Always use the dictionary especially if you're working with the vocabulary Double check with the reputable sources

[13:03] If you're learning English that's Merriam Webster Cambridge Oxford Language Oxford Learners or Oxford

[13:11] Advanced Learners Dictionary Everything is available online There is also a good tool like Context Reverse

[13:18] And these are all my personal recommendations They have not like I'm not endorsing them or anything or promoting them

[13:25] These are just recommendations based on the experience So always verify the output and use the dictionary to verify for example vocabulary

[13:37] Or even if you are learning grammar always make sure that you verify against a reputable book or printed source

[13:47] I've had this problem when I was learning le subjonctif It's like a specific mood in French And yeah

[13:58] AI also hallucinated and made a mistake And the explanation was almost wrong almost completely wrong

[14:05] And I had to I was because I do have a bit of background knowledge I had had background knowledge even before studying this topic because

[14:14] I had a very good tutor So she already had explained this to me And I was like somehow trying to recap to do this revision

[14:23] And so I asked AI and that was back even before I did some training you know in IT Like I understood the true value of getting a paid version of

[14:37] LLM model And so yeah it made a mistake and I had to use the grammar reference book And so the last thing

[14:46] I want to note is that and this is like it's kind of tied to what I was saying previously Is that try to work on your prompt make it very specific

[15:00] And also try to give some PDF files or markdown files that are even better as your input Like I said your markdown file can be just the

[15:11] Google document lesson that you had with your teacher Let's say like this But all in all I got to say that

[15:17] I do use it all the time I think it's a lifesaver when it comes to consulting and like helping me with checking written works of my students as well

[15:30] But I always verify Like of course it is helpful It helps me a lot but I always verify And the only thing and the only way

[15:40] I can verify is that based on my experience on my expertise as a professional as an educator And also double checking with the reputable sources like grammar reference books vocabulary reference books

[15:54] And I will even create I think a separate episode about that If you're learning English or French I will make the list of my personal favorites and

[16:05] I'll make an episode about that So stay tuned And I just want to say that we got to embrace AI and language large language models for sure

[16:16] Try playing around and buying the paid versions Try working on your prompt engineering writing prompts

[16:25] My husband even said you know what Like since the AI and all the LLMs came to use he said oh you know my

[16:36] English has improved And he's not an English teacher He's in the IT field right But he said that I write so many prompts every day at work

[16:45] So it helped him to improve his writing actually because you got to think about the prompt You got to think what you're asking

[16:53] And you know I use AI when I'm even like not sure during my study buddy sessions As I've been as I told you guys before in the previous episodes

[17:06] I have study buddy sessions with some people who learn French with me So we don't pay each other We just study together

[17:16] And so sometimes we have some doubts So we ask my paid version of Claude and I just log in and I have the separate project

[17:25] It's very nice because you can spread your chats into projects So let's say you have the project big project of

[17:36] French learning right And you will stack all the necessary PDF or markdown files in that project And you want to provide more context

[17:45] And you want to make sure that the memory mode is zoned because in that way it remembers everything about you all the background everything

[17:54] And I do find it useful especially if you're using for personal reasons And it's not like yeah it's useful when it remembers a lot about you

[18:06] It's not dangerous as long as you don't provide anything super sensitive So you've got to be smart about it

[18:12] Of course the responsibility is always on you what you want to provide So I'm not advising and saying that you know

[18:20] AI will or LLMs will replace a teacher No it's not like that I'm this kind of person who is both a teacher and a student because

[18:28] I am an English teacher English is a second language teacher And also I keep learning French and I want to master

[18:35] French really well before I actually start teaching probably So I'm just trying to point out that I do think that

[18:46] AI and LLMs are not going to be able to replace an actual human teacher because language is all about spontaneous interaction

[18:54] And LLMs cannot give us that So for now they cannot But I would say most importantly it should always be about combining

[19:03] LLMs and the teacher And in that way you can benefit greatly Thank you so much for hanging out with me today

[19:12] And stick around and hang out a bit more In the next episodes I'll be talking a bit more about my recommendations when it comes to

[19:22] English and French learning So I'm sure you're going to benefit a lot from that And also I'm preparing like an episode just recapping and talking about all kinds of things that have been happening recently because

[19:38] I've been really busy lately So yeah So just stay tuned and have a great day See you next time Bye bye