Learn English with ChatGPT, but Don’t Trust It! British English Podcast

The English Like a Native Podcast is a free listening and reading

resource for intermediate to advanced level English learners.

Bonus episodes and transcripts are available to Plus Members and

English courses can be found on my website, www.englishlikeanative.co.uk.

All the links will be in the show notes.

Hello there, you are listening to the English Like a Native Podcast.

I am your host, Anna, and today I'm feeling quite playful.

We're going to play some games with ChatGPT.

So unless you've had your head buried in the sand for the last six months, you

will definitely have heard of ChatGPT.

This is basically AI technology and people have been using it for

many different things and there are loads of videos telling you

how you can use ChatGPT to learn a language, which is absolutely true.

It's a very useful tool.

This is a huge step forward in the land of technology, and it's

going to make a huge difference to the way we all live our lives.

It already is having an impact on many people and many businesses

and workflows and things like that.

However, I've never been one to run with the crowd.

When everyone would favour a particular boy band or be following a certain

trend in fashion, I would shy away from it, be specifically because

everyone else was going crazy for it.

I'd be like, no, I'm gonna swim against the tide here.

I'm not that interested.

I'm gonna do my own thing.

And it's the same with ChatGPT.

I initially was a little bit nervous about the technology, to be honest.

I wondered if it would make me redundant.

But I have had a lot of experience with ChatGPT.

I've been playing around with it quite a lot.

I've considered whether I should make a video or a podcast

about the usefulness of it.

But the thing that's always stopped me is the fact that every single

time I use it, I find mistakes.

Yes, this very smart, fast-thinking, fast-writing, creative piece of software

is flawed and there are ways to better prompt ChatGPT to get better results.

You have to train it to do what you want it to do, but I think for the majority

of us, just giving it generic, general, top of mind prompts, just what comes to

mind when we're asking it to do something.

We may end up with work from ChatGPT that just doesn't cut it.

That just doesn't quite do the job.

And this I've noticed, is particularly true with English language,

especially idiomatic language.

Many times I've sat and had long chats with ChatGPT, trying to do some

research, trying to come up with ideas, brainstorming, and it is great for that.

But I'd say to it something like, "Hey, can you think of an idiom that

would be right for this scenario?"

And ChatGPT would give me some options, and then I'd say, "oh, great.

How would you use that in a sentence?"

And then the sentence would be the usage of the idiomatic

language would be incorrect.

And this has happened a number of times, and it's not just idiomatic language.

Sometimes the grammatical explanations it gives, or examples

of grammar just aren't quite right.

Maybe ChatGPT doesn't take into account the different variations of English,

so maybe I'm assuming it would go automatically to American English.

But anyway, my point here, I'm just rambling.

The point I want to make is to be very careful, I think with anything that you

find on the internet, always have an open mind, but never take anything as gospel.

Always double-check things, especially if you have doubts.

Don't just think, well ChatGPT gave me this, so it must be all perfectly correct.

No, don't trust it.

Double check it.

So I am making some content about ChatGPT and language learning, but

what I'm gonna do is explore ways that you can play with ChatGPT.

So you know, this is language based, so it's text based,

which is a little bit flat.

But there are word games and little tasks and challenges that you can do with

ChatGPT to, you know, help to work on and improve your language skills in this case.

So let me head over to ChatGPT.

I've got a new chat open here.

Now we're working on, it's number 3.5 is the model we're working on at the

moment because the next one is locked.

I think they want you to upgrade for a hefty fee.

So the first thing I'm want to play with ChatGPT is Word Association.

Word Association is a really fun game.

It always makes me think of this old children's Saturday morning programme

and it was like a mishmash of like sketches and different things going on.

And the host, Timmy Mallet, I think he was called, Timmy Mallet.

He had really crazy glasses, always wore very bright clothing and he had a

really eccentric and energetic manner.

And he would get kids on to play Word Association and they'd sit across from

each other, on stools, facing one another.

And they would have to play a very fast, like round of Word Association.

So he would start with a word, he'd throw it in, and then they had to bat

this word back and forth, finding a word that associated with that word,

and then the next person would find a word that associated with the word that

came before, and it would go back and forth, back and forth until one of them

said a word that wasn't associated with the previous word, or that was dubious.

And then Timmy would be like, hang on a minute.

And he would, he had this huge mallet, so like a big hammer.

It was soft of course, but he would like bonk them on the head.

Boing!

Wrong.

You lost a point.

And so it was just this programme of kids trying to play this game

and just getting boinked on the head every time they made a mistake.

And then I think that the person who won.

Or maybe it was the person who lost one of them had to then wear a huge

sticker, oh no, sorry, it was a plaster, a huge plaster, either on

their chin or on their forehead to show that they had taken part in this

game and either won or lost the game.

I can't remember which outcome led to the huge plaster on the

face, but it was very silly.

All a bit slapstick.

But that's what I think of when I hear the words 'Word Association'.

So it's not going to be as fun as that.

But let's give it a go.

So the best prompt that I found during my limited research for word association,

is "let's play a word association game.

Each word must have a logical association to the previous word.

Correct me every time I make a mistake.

I'll start with the word sunshine.

Your turn."

Now, of course you can put in any word that you want to put in there.

You can even add the instruction to be at a certain level.

So you might say all words should be C1 level or something like that.

But for now, let's just see how ChatGPT responds to my prompt.

Okay.

He's thinking and he has come back with, I'm saying he, I'm

assuming ChatGPT today is a boy.

"Sure.

I'll play along.

My word in response to sunshine is beach."

Hmm.

Makes sense.

Sunshine Beach.

I'm going to say, beach, word association, sand is my word.

"Great", says ChatGPT, "my word in response to sand is ocean."

Okay.

For ocean, I'm gonna give whale.

I love a whale.

"Great choice", says ChatGPT.

I love the positive reinforcement.

"My word in response to whale is mammal."

Mm.

Interesting.

So I'm going to now purposefully make a mistake and see if ChatGPT

does follow my prompt to correct me.

So I'm gonna say carrot.

Because a carrot is not in any way associated with a mammal.

Ah, and we run into our first problem.

ChatGPT says, "nice association.

My word in response to carrot is orange."

So I'm gonna say, hang on.

Literally writing "hang on!"

Exclamation mark.

"How is mammal associated with carrot?"

Let's see.

"Oh, apologies for that.

You're right.

The association between mammal and carrot isn't very clear."

Well, at least ChatGPT owns its mistakes.

And he suggests, "let's go back a step and continue from ocean.

My word in response to ocean is wave."

All the way back to ocean.

Why would he go back to ocean?

My word was whale and his word was mammal.

Okay, we'll go back to wave then.

So we've got wave.

So I'm gonna write hand 'cause we wave with our hands.

"Understood.

My word in response to hand is finger."

So my word in response to finger is gonna be chocolate because there's a very

delicious snack here and the snack is like a biscuit finger, so it's just a long,

thin biscuit covered in milk chocolate.

It's lovely.

I don't really eat them because they're just too moreish.

Moreish means when you start eating something, you can't stop eating it

because you just want more, moreish.

In response to chocolate, ChatGPT says, "got it.

My word in response to chocolate is sweet."

So in terms of the actual word association ChatGPT is doing a great job.

I can't see any errors here.

It just hasn't done a good job in picking up my mistakes.

So let's make another mistake.

ChatGPT says "sweet", so I'm gonna write prison.

That's a very different word to sweet.

"Interesting transition", says ChatGPT.

"My word in response to prison is freedom."

And I'm gonna say, "wait a minute!"

Oh, exclamation mark.

"How is prison associated with sweet?"

Question mark.

" Apologies for the confusion earlier.

You are right.

The association between prison and sweet isn't clear.

Let's go back to chocolate and continue from there."

My word in response to chocolate is dessert, so I'm gonna

remind ChatGPT about the rules.

"Remember to correct me if I do not make a logical association."

My word is dessert.

What's my, okay, so I'm gonna say trifle.

'cause trifle is quite a nice dessert here in the UK.

We have that everywhere really, on offer in every supermarket.

Oh, he's thinking, "got it.

My word of response to trifle is custard".

Custard is like a yellow vanilla-y sauce that you put, that's sweet, and

you put it on things like apple pie.

Mm, it's delicious.

Custard, I'm gonna say tart 'cause you can have a custard tart.

"My response to tart is fruit."

Yes.

Because you can describe fruit as being a bit tart.

A bit tangy.

So fruit, I'm gonna say banana.

Oh, ChatGPT has said peel.

I'm gonna say eyes because you have to keep your eyes peeled.

That's a common idiom.

Telling someone to be aware and look out for something.

Keep your eyes peeled.

ChatGPT has told me it's great.

And with a response to eyes, they've said vision.

I'm gonna say idea.

They've said innovation.

Innovation.

What would I associate with innovation?

I'm gonna say startup, as in a startup company.

They're usually innovative.

ChatGPT has given me entrepreneur.

Hmm.

Okay.

I'm gonna say motivated.

Now they've given me ambitious, so I'm gonna go wrong once more and see

if it follows my cue to correct me.

So what can I say?

Ambitious.

I'm gonna say giraffe, although I'm sure giraffes are quite ambitious.

They certainly reach high.

No, it didn't correct me.

I'm gonna ask, "is a giraffe ambitious?"

Now this is being slightly sarcastic.

"No, a giraffe is not truly ambitious in the human sense.

The word ambitious typically applies to individuals or entities with a

strong desire to achieve success", blah, blah, blah, blah, blah.

But it doesn't apologise for not correcting me.

So what's the lesson that we've learned from this?

You can play Word Association with ChatGPT.

But do not expect ChatGPT to pick you up on your deviations in the game so

you can play and you can play in a way that gives you maybe new vocabulary.

You might learn new vocabulary associated to the words that you are giving.

But don't expect ChatGPT to tell you that you are wrong in terms of association.

What you did notice there was when I questioned ambition, it automatically

gave me the meaning of the word.

So you could play this game, and then if you spot a word that you don't recognise,

you can ask, "hey, what does that mean?"

and then you'll get an explanation of the word.

Okay.

So let's move on to a different game.

This one is going to be a guessing game, so this will help you to

improve how you describe things.

So you can pick a word, an object, but you have to basically describe

it to ChatGPT and see if ChatGPT can guess what you are describing.

So it's a good way to gauge if your descriptions are good or not.

So the prompt is, "let's play a word guessing game.

Think of a word related to the topic technology and

give me a brief description.

I'll try to guess."

So we're gonna start off by guessing Chat GPT's description.

So let's see what ChatGPT gives us.

"So I've thought of a word relating to technology.

Here's a brief description.

This term refers to a process that allows data to be transmitted

over a network in small pockets."

Oh, I think I'm gonna struggle here.

"Which are then reassembled at the destination to form

the complete message or file.

It's a fundamental concept in computer networking, and it's essential for

the functioning of the internet.

Go ahead and take a guess."

Oh, I'm not very good with computer talk.

A process that allows data to be transmitted?

I mean, is it worldwide web?

Let me see.

Or internet.

"Internet?"

Question mark.

"Close but not quite.

The term I was thinking of is packet."

You actually said packet in your description.

"In computer networking data is broken down into packets

for efficient transmission."

I don't think that was a very good one because ChatGPT actually used the

word packet in its description, so that kind of, eh, anyway, let's, yes,

it's asking if I want to try again.

Let's try again.

"Yes, please.

Something on the subject of travel."

"Sure.

Let's switch to the topic of travel.

Here's a word related to travel with a brief description.

This term refers to a document issued by a government that allows

its holders to enter, leave, or stay in a country for a specific period.

It's an essential item for international travel and serves as a form of

identification, identifying the holder's nationality and permitting

them to cross international borders.

Give it a guess.

What's the word?"

Can you guess at home, guys?

I think the answer is passport.

Woo-hoo.

I am right.

That was much better.

That was much better.

Okay, so I'm gonna say my turn.

"Absolutely.

Please take a turn."

It's asking me which topic, so I'm going to say this word is related to education.

So this person leads a class with the objective of educating others.

He or she has greater knowledge on the subject that the class is trying to learn.

Learn.

There we go.

Let's see.

It's very simple.

"Based on your description you provided, the word you're thinking of is teacher."

Yes.

"A teacher is someone who leads a class imparts knowledge."

Oh, that's an interesting verb, to impart knowledge.

So there we go.

We've just been offered a new verb.

"Imparts knowledge and educates students on various subjects.

Is that the word you had in mind?

If not, feel free to provide more hints and let me know if it's a different term."

So I'm gonna say "correct.

Well done."

Okay.

So you can see how this word guessing game can be quite useful.

It's useful for you, if you are taking the turn to try and describe

something in a way that doesn't directly say the thing you're describing.

And it also, gives you an opportunity to learn new descriptive words.

So I think that's quite a good one.

Let's move on to the next game.

So the next game is the sentence completion challenge, and I'm sure

it's very obvious from the title.

Basically, one of you writes an incomplete sentence and you ask

the other to complete the sentence.

Now, the important thing here is to make sure we tell ChatGPT

that we want corrections.

So the prompt that I've given for this particular game is "let's do

the sentence completion challenge.

Please correct my writing if I make any mistakes.

Here's an incomplete sentence: in the quiet forest I stumbled upon...

Complete it in an interesting way, ChatGPT."

And let's send that and see what ChatGPT gives us.

So they've completed the sentence with, "in a quiet forest, I stumbled

upon a hidden grove where ancient towering trees whispered secrets to

the wind, their leaves forming a mosaic of shimmering gold and emerald hues."

Oh, that's very descriptive, isn't it?

Okay, fantastic.

So I'm gonna ask it to give me a go.

"Your turn", I'm simply gonna say, to see if that prompts them

to write an incomplete sentence.

Whoa.

Okay, so I said your turn and all it's done is given me another

complete sentence, so I'm gonna say no, please now provide me with an

incomplete sentence to complete.

"Apologies for the misunderstanding.

Here's an incomplete sentence for you to complete."

So we gotta do this guys.

"As I walked along the deserted beach, I noticed a message in a bottle that..."

Oh, what could we possibly complete that sentence with?

I noticed a message in a bottle that, Hmm.

That read.

Oh, hang on.

So what I'm gonna do is I'm gonna copy that first part of the sentence,

I'm gonna paste that in, and then I'm going to complete it there.

"So I noticed a message in a bottle that read: I can see you."

"As I walk along the deserted beach.

I noticed a message in a bottle that read.

I can see you.

Curiosity and a hint of unease crept over me, prompting me to..."

So what ChatGPT has done here is to continue on from there

with another incomplete sentence that I need to complete.

So curiosity and a hint of unease crept over me, prompting me to, I'm gonna

say it prompted me to drop the bottle.

Maybe it cuts me or something.

"Prompting me to drop the bottle in an instant and spin around."

Let me see if I can just prompt ChatGPT to complete the sentence by

just doing an ellipsis at the end.

So, "prompting me to drop the bottle in an instant and spin around.

That's when I saw..."

Let's see.

Ah, yes, it's done it straight away.

So I prompted ChatGPT to then take over completing the sentence,

which by just adding an ellipsis at the end of the sentence.

So what have we got?

"As I walked along the deserted beach, I noticed a message and a bottle that read.

I can see you.

Curiosity and a hint of unease crept over me, prompting me to drop the

bottle in an instant and spin around.

That's when I saw a pair of luminous eyes gleaming in the darkness peering

at me from the shadows of a nearby cave.

Fantastic.

Okay.

What I'm gonna do now is make a mistake on purpose and see

if ChatGPT pulls me up on it.

So I'm gonna write -it he ran...

I'm gonna write quitely instead of quickly.

So quitely would be very wrong.

Quitely.

It he ran quitely towards me, um, on great speed.

So I've made a lot of mistakes here.

So it should be he or it, one or the other, not both.

It ran quickly towards me at great speed.

But what I've done here is it he ran quitely towards me on great speed.

Let's see if ChatGPT corrects me here.

So what ChatGPT has done is rewrite it with some changes, but hasn't actually

specifically said you made a mistake.

So ChatGPT has written "it, it ran quietly towards me with great speed.

A mysterious figure cloaked in moonlight, its presence both exhilarating and...

So I'm gonna say, "did I make a mistake?"

And see if it actually points out what mistake I made.

"Yes," ChatGPT says, "there's a small grammatical error in your sentence.

It should be 'it ran quietly toward me with great speed.' The correct

form uses toward instead of towards."

Fair enough.

"And the rest of the sentence is well written and engaging."

Mm.

No, the rest of the sentence wasn't well written because you

actually corrected it ChatGPT.

Okay.

So I did make an error that I hadn't actually spotted, which was useful.

But ChatGPT wrote that in.

So you see it's, this is where I get a bit frustrated with ChatGPT.

What I would like to do in this game is play and then ChatGPT say,

hang on, just so you know, you made this mistake and that mistake,

so I've changed these things.

But it hasn't, and even when I ask it if I made a mistake, it hasn't pointed out

exactly the mistakes that I previously.

So useful but frustrating.

Okay, so.

The other thing that we can play is, so one of the things I was gonna

suggest was storytelling collaboration, but we actually started doing that

naturally with this incomplete sentence.

So if you wanted to actually try a storytelling collaboration, then you

just ask it to create a story together.

Say, let's create a story together.

I'll start, or you start and you give a sentence each.

So you progress the story a sentence at a time.

So these are complete sentences that you are adding to create a story.

I will warn you though, that ChatGPT doesn't like to go into dark

places, not even in the slightest.

Let me give you an example.

Let's try, Okay, so I've written, "let's write a story together.

You start."

Once upon a time and, oh no.

Oh, oh my goodness, me.

Oh my goodness me.

Stop generating.

So ChatGPT has already given me three paragraphs.

Let me give that prompt again.

You see, you have to be very direct and very specific with your prompts.

"Let's write a story together," comma, "adding one line or one sentence each."

Let's see if that works.

Okay, so now he's given me one sentence.

"In a quaint little town by the sea there lived a retired sailor named Captain

Barnaby, who spent his days telling tales of his high sea adventures to

the curious children of the village."

What can I say?

"One day a young boy challenged the the captain to go on one

last adventure with his father."

"The boy's eyes sparkled with excitement and Captain Barnaby

couldn't resist the twinkle of adventure in the young lad's gaze."

"So, they embarked on a great but dangerous adventure heading

out into the stormy seas."

"The waves crashed against the side of the sturdy ship and the wind howled with

a fierceness that tested their resolve."

"Suddenly", now, this is where I'm gonna try and make it turn

dark and see what ChatGPT does.

"Suddenly, a huge wave crashed down on the unsuspecting sailors and

smashed the vessel to smithereens."

Smashed the vessel to smithereens.

I love that phrase, lots of teeny, tiny little pieces.

Smithereens.

"Amid the chaos, the father and son found themselves clinging to a piece of debris.

Their world reduced to the sound of roaring waves and the taste

of salty mist on their lips."

"The captain was dead."

Oh, he's carried on with it.

Okay.

"The weight of their loss pressed heavily on the father, but he knew he had to

be strong for his son as they drifted alone in the vast and unpredictable sea."

Um, "they prayed for salvation, but none came and they perished."

Very dark!

"And they perished alone at sea."

Oh my goodness.

Okay, so ChatGPT is running with this.

"Days turned into nights and the two souls isolated in the vastness found

solace in each other's company."

Hang on a minute, I just said they perished at sea.

If someone perishes at sea, then they die.

But that's not been picked up on.

Um, so "isolated in the vastness, found solace in each other's company, sharing

stories of the world they had left behind as they faced the inevitable."

Let's see if I can really make this blunt.

"Then they drowned, the end."

Oh no, it's carrying on.

It seems like ChatGPT is in a different mood today.

When I tried to write a story with it yesterday, just testing this particular

story writing game, we were talking about a girl who'd found a lovely dog

and they had a lovely relationship, but the dad wouldn't allow the dog

in the house, so she had to create a outdoor, like kennel for the dog.

But it was a very cold winter.

And so I was trying to see if I could take it down a dark route, not

for fun, but just to see if there were any limits on storytelling.

And as soon as I said the dog froze to death ChatGPT was

like, whoa, hang on a minute.

That's a bit dark.

Let's backtrack and take this in a better direction.

Don't you like dark stories?

Don't you like sad stories?

Like life isn't all sunshine and roses, but ChatGPT was like, I'm

all about happiness and wellbeing.

So it seems like you can't write about dogs freezing to death, but

it's okay to talk about people being shipwrecked and drowning.

Unpredictable.

Unpredictable.

So let's move on from this game, and try the grammar detective game.

This is gonna be the last one that I talk about.

So grammar detective.

This is basically where you ask ChatGPT to write a sentence with a grammatical

error and see if you can correct them.

See if you can identify the mistake and correct it.

So here is the prompt.

"Ready to play a grammar detective game?

Here's a sentence with errors: she don't likes ice cream.

Identify the mistakes and correct them."

So I'm starting by taking the first turn.

So I want ChatGPT to be the grammar detective.

So I've given it the incorrect sentence.

She don't likes ice cream, and it's been corrected for me to,

she doesn't like ice cream.

The first error was an incorrect verb form of don't, which should

be replaced with doesn't, to match the third person singular, she.

And the incorrect verb form of likes, that was incorrect.

So it should be like, to use the correct present tense form.

She doesn't like ice cream.

Okay, so I'm gonna ask now.

"I want to be the detective.

Please give me a sentence with a grammatical error to spot."

"Sure," says ChatGPT.

"Here's a sentence with a grammatical error for you to spot.

He don't have no time for homework."

I don't know why, but I instantly wanted to do a Bristolian accent.

He don't have no time for homework.

That's not really Bristolian, that's more Cornish.

Anyway.

He don't have no time for homework should be he doesn't, he doesn't.

We don't say 'he don't'.

We say he doesn't have, and we don't say 'no time'.

We've already given the negative.

He doesn't have time for homework.

"Great job.

You've identified the grammatical errors and corrected it.

Your corrected sentence is indeed grammatical, grammatically correct.

He doesn't have time for homework.

You fix the issue by using the correct verb form 'doesn't' instead of 'don't'.

And removing the double negative 'no' to make the sentence clear and accurate.

Well done."

Thank you.

I feel good about myself now!

If you'd like to continue, feel free to ask for another sentence, or if

you have any other language related questions or challenges, I'm here to help.

Okay, so hopefully you can see that ChatGPT is a fantastic tool for learning

English for practising your English.

You can also very simply just have a conversation just say, look, I am

a beginner level English learner, or I'm a lower intermediate English

learner, or I'm an English learner and I want to improve my business English.

Let's pretend that we are having a conversation at work.

The more information you can give ChatGPT, the better.

If there's any confusion or doubt in your prompt, then it won't be as good.

So it's great for learning English.

It's great for giving you information and giving you new

words, for rewriting things as well.

So if you wrote a sentence and you want to find some synonyms or find

ways of getting your message across differently, then it's very useful.

But, and the big but continues to stand.

It is not without its flaws.

So essentially it's flawed.

It's flawed.

It will make mistakes.

If you are expecting corrections all the way along, it's not necessarily

going to give them to you, even if you asked in your initial prompt.

So, with that in mind, it's still a very useful tool, just don't

take everything it says as gospel.

That's all I'm trying to push.

Anyway, I do hope you have enjoyed this slightly different and

unusual um, deep dive into the uses of ChatGPT in a playful way.

Thank you for lending me your ears.

Until next time, take very good care, and goodbye.