ChatGPT for translating, correcting and improving texts

 Today we'll dive into use case of using ChatGPT to translate emails from English to German or any other language, if you will. And not only that, we'll also try to correct them for grammar in style.

Hi, I'm Christian Principle Project Measure at appliedAI. Welcome back to the video series, how to Master AI Tools at Work. This video is designed for people who speak two languages, a native language, and another language where they're not super fluent in.  We will dive into three use cases, translating emails from one language to the other, then correcting them, and then improving the emails for style.

And at the end of the video, we'll concatenate all of these steps into a singular problem.

Okay, head over to the OpenAI website. And what you find on this website is this sign up button, click it and follow the process.  The first thing that the system prompts is your email address. So I just enter my email address in here.  and click the continue button.  I then have to provide the system a password.

Let's use a very simple one for now, but always make sure that you have a very secure password. Let's just go for this one.

Okay.  That's the password. I hit continue.

And then ChatGPT asks you to verify your email address to make sure it's valid. So jump over to the email address itself. I look into my inbox.  In my email provider, I can find the email here, my junk email, so always double check if it's in the spam folder or not.  Then I have to verify in my email, I click it  and I'm being redirected to the T webpage.

Of course, I do have a name that would be Christian  and a birthday, and I hit agree.

And boom, that's it. We've signed up for ChatGPT. And now you can use the system and become creative and use it for all kinds of use cases.

So let's start with email translation. For that we will use the so called RTF framework, which stands for role, task and format. This is one of many prompt engineering techniques. If you're interested in that concept, just have a look at this video here above.  So, RTF stands for Role, Task, and Format, and it basically says that if we prompt ChatGPT, we assign it a specific role, we tell it what task it should fulfill, and then we specifically tell the system what kind of output it should generate.

Let's have a look at ChatGPT to see how that looks in action. Here's the prompt. We say, You are.  An email assistant.  Clearly, this is the role that we give to the system. Jetty PT now is an email assistant. It should help us in drafting this email, translating it. Next, we assign the task to the prompted, say, translate the following email in German, in English.

to German. That's the task. And what's missing is the format, what kind of format the output should be. And he would say, basically, it's text. Could be other things, could be a list, could be a table. But for now, we're happy with a text because an email is basically just a text. Okay, I hit enter. And then I get a confirmation from the system that indeed, it understood what I'm looking for.

Next we need the email. So here's a, an exemplary email. This is just an email to a person called Susanne. It has a subject line and also a name in the end. It's just a typical email. It's in English. I send the message and then it instantly get the translated email in German.  So that's it. Pretty easy, isn't it?

It might be that you're not super fluent in German. You know the language, but in a very proficient and sometimes you don't catch specific errors. So let's use ggpt to catch those errors for us.  Here's the prompt we will use.  Now we're telling the system that, again, you are a mail assistant.  And correct, here's the task, correct the following email  written in German  for language errors.

Then we have the task ready and then we need the output again. Again, this is just a piece of text. Format the output  as text.

Naturally, the system doesn't do anything. It just confirmed that it understood the prompt and now we need to assign the email itself to the system. So that's the email. Let's have a closer look at the email before we hit the button.  Here we have a few typos in here. If we look closely, for example, we see the word Konzeption, which is just a German word for conception, but it's obviously wrong because it has this umlaut in here.

And we also see other problems. For example, this word is missing an n in the end. This should be abzuschließen instead of abzuschließen.  So a few typos in here. So let's see what ChatGPT does if we hit send message.

Looking at it again, we can see that these two mistakes have been corrected. For example, Konzeption now is The right word without an umlaut. And also we can see that this word abzuschließen ends with an N. So it corrected all the mistakes. If you want to, you can also prompt ChatGPT to create a table or some other format where you can specifically inspect the different errors and problems.

We can briefly do that.  So let's say,  please create a table to visualize  the mistakes.  In the email  And what we get is a list of all the mistakes we've made in the original email and also the corrections as well.

If you're not proficient in a specific language You might miss the subtleties of the language in particular in a professional setting you might say a few words that are not Perfect for the kinds of setting you're working in and for that we might need a little bit of help with ChatGPT So let's see that in action  In ChatGPT, I will open a new chat.

Why? Because ChatGPT remembers the previous conversation and it might correct the email I've just written, but now I want to correct another email for style. So here's the new chat. And as always, I need a prompt and here's my prompt.  You are a writing coach.  We have the role. Then we need the task, enhance the clarity and professionalism  of the following email.

Written in German.

And then the format itself, which is again text. Format the output  as text. That's it. We hit enter. And again we get a confirmation that it understood the prompt.  Here's the email. And let's have a closer look again.  So this email is very informal. It's written in German and you might catch a few things that aren't not super professional.

For example, in the beginning, it says, not something you would usually write in a professional context, or it also has this word marketing ding, which is something you might say to your friends, but not really in an email. Okay. So a few things that we need to correct for when it comes to style. Let's hit send message and see what happens.

Clearly, this is way more formal now. It starts with this  Sehr geehrte Susanne. So instead of just saying hi Susanne It writes in a more formal way and also corrected some other problems in terms of style and instead of saying hey  What was it? Alles klar bei dir? It says instead ich So a much more formal and better email which you can send out to people in a more professional context You So that's a wrap.

Three tasks, and it's very easy to do them with ChatGPT.

So far so good. But we can be even more effective if we use a technique called prompt chaining. With prompt chaining, we write a singular prompt with many steps and do everything we just did in one go. Let's give it a shot.  Again, I open a new chat. So  ChatGPT doesn't have any history of what happened just before.

And here's the prompt. Now we're you are.  a language expert. So make it more broad, the role itself.  First,  correct the language  of the following email draft  I will give you.  So that's the first step. Now comes the second. When the text is corrected,  translate  the email  from English to German.  And finally, last step, enhance the professionalism  of the email.  So now we have written the task, but the task is just a sequence of steps.  And then we need the format itself in the end. Format  the output  as a text in German.

Nice. So let's give it a shot. Send message.  And again, it asks me for the email.  Here's the email.  It's not the perfect email. It has a few mistakes. For example, here there's a space missing between quickly and tell. It's also slightly informal. Just a good email for this prompt.  Let's hit send message and see what we get as an output.

So what does it say?  It creates the email in English as well with improved grammar and style, but it also gives me the same email in German, again, with a good grammar and a good style as well, which is more formal and applicable to a professional setting.

At appliedAI, we take Trustworthy AI very seriously. And when it comes to using ChatGPT, there are a few things to consider. And the first thing is how your data is being treated by OpenAI. Because by default, your data is saved and stored, so everything you type into ChatGPT will be used by default by OpenAI to improve ChatGPT as a service and their models.

You might want to disable this feature. That's how it works in ChatGPT. Here on your name Here, this is me. There's the setting menu and if you click it, there's the data controls menu and here you can deselect the chat history and training feature. If you deselect it, your data won't be used anymore for model training and model improvement.

But keep in mind that you have to deselect that feature every time you log in to ChatGPT from a different browser or a different system.  It doesn't change it across your devices. So that's something to consider. The other thing to consider is that OpenAI has a 30 day retention policy, which means that even if you disable this feature, they will store your data for 30 days for monitoring abuse, for example, like criminal conduct on your site.

There's way more to cover when it comes to trustworthy AI and the legal issues. And if you're interested in that stuff, click the link below this video. Well, the application of ChatGPT poses a few risks. Let's talk about them.  Basically what you have to know is that ChatGPT doesn't have any clue about your context in which you operate.

It doesn't know which industry you're working, it doesn't know the department, it doesn't know which kind of company you're working. And that leads to a few problems and risks. The first problem is accuracy. Some documents are very specific. Technical reports are medical documents, for example. And for that, you need a very specific language.

ChatGPT might not catch the right terms at times and use different terms that are not used in your context, really. The other problem is a loss of nuance. Your department, your team, has a specific tone, has a way of talking, of communicating.  And. If ChatGPT uses a different kind of communication style, that leads to some problems and unintended consequences.

Maybe some people take offense with the kinds of things that ChatGPT generates. And then we have the problem of confidential information. Sometimes emails contain sensitive information, such as emails or names from other people, and you will ingest that information into ChatGPT, and ChatGPT might use it to train and improve the model itself.

It happened before that. Emails were teased out of the system that were written by someone else in the past. So the key message for you as a user of ChatGPT is that please double check.  Don't leave your expertise at the doorstep rather have a look at the output once more to see if it's really valid, if it applies to your context and only then use it and send it out to other people.

Well, ChatGPT is just one of many tools you can use. There's also BART by Google or Bing by Microsoft or Llama by Meta. Choose the tool that you're most comfortable with. That's a wrap. We finished the video and I hope you found it helpful. You can use the information we talked about today and if you like the content, hit the like button, subscribe and hopefully see you next time.

Bye bye.