How to write perfect prompts for ChatGPT

This image has an empty alt attribute. The file name is itunes_badge_md-300x110.png

Introduction

Nice to have you with us again for another episode of Digital for Productivity. After we went through the prompting recipe in the last episode, where we said: How to create an optimal prompt for ChatGPT, in this episode let’s have a few more tips. If you haven’t heard the episode with the prompting recipe yet, here it is in short form. The recipe stands for six tips for the optimal prompt, “R” stands for the role, “E” for the result you want to achieve, “Z” for the target group you want to reach, “E” for the place of publication, “P” for samples and “T” for tone. More in the episode on the perfect ChatGPT prompting recipe. More prompting tips in this episode.

First tip: Ideally, you should proceed step by step. What does that mean? If you are writing an article, for example, then in the real world, if you are doing this with an employee, you should also proceed in such a way that you first entrust them with the task: Why don’t you collect a few topics? Then they come back with topics and then you say: “Gosh, please suggest a few titles for the topic we’ve chosen together.

Once you have the title, then you make the subtitle, then you make an outline and then each paragraph. And you should proceed in exactly the same way with ChatGPT. You’ll get much better results than if you say: Do you write the article on this topic? You usually get better results with real employees too. With employees, it’s often the case that if you delegate this to someone more often, then you can take steps out of the process and you can also do this with ChatGPT, for example, if you build your own GPT – this is possible in the paid version – and then you can also pre-structure this iterative process and say: Okay, create this first, then this, then this. So what an employee learns, they can of course also set up there by writing a GPT. The second thing is to formulate it positively. So you say: don’t write this and this. It’s the same if you have children who say: Don’t run into the street. The brain has a hard time with negations. That’s the classic: don’t think of a white elephant, and bang, you’ve got the white elephant in your head.

And artificial intelligence also finds it easier if we don’t confuse it. Ideally, you should formulate things positively and not use too many nested words that make it difficult to determine what is being negated or what is not being negated. So formulating it positively, briefly and succinctly will give you better results. The third point: A very good strategy is to have it explained to a twelve-year-old or a five-year-old. This means that we often have to write very, very complicated, long boxes and the AI is trained on the texts that we have produced as adults. And when they then say: Either you’re a great copywriter, Jung von Matt for example, and you can say: This is how you orient yourself to the Sixt ads. That’s one approach. The second approach is to have it explained for a younger target group. They often get much clearer statements and texts that can also be used for older groups. Then you should also make sure that your answer is unbiased and not based on stereotypes. In other words, you have to be careful here. For example, it is now the case that AI-generated images for the topic of gender neutrality and so on also show female soldiers, including some with black skin, for example, which no longer necessarily corresponds to the image of the time.

This makes it all the more important to be very clear about what is correct right now, but to be as neutral as possible and always focus on the result you want, just as you would if you were writing to an employee. Then always ask questions. You can also tell ChatGPT: So, please tell me when you have understood it, give me an okay, otherwise ask me a question. You could do the same to an employee. I learned this in one of my first leadership training courses, to say: When you delegate something to someone, please have the employee repeat back to you what task you gave them. I know this from my time in the army. It was also an issue there to make sure that it was received correctly. Ideally, you should then prompt in English, as most of the training data is generated in English. This means that the AI is much better in English. Of course, you can also prompt it in German. The only difference is that, as a rule, your prompt is first translated from English into German, processed in English and the result is then translated from English back into German.

This means you have two translation processes and, of course, something can always go wrong with each translation process. If you are not so good at English, then I have two starting points. The first is to learn English. If you can speak English today, you’ll be out of a job tomorrow. Sorry, that sounds a bit harsh, but I’ll stick to saying English is simply extremely important professional communication, like writing and arithmetic, so against that background. The only excuse I’ll accept is if you say you speak Spanish and Chinese, then you can let it slide a bit with English, but ultimately Spanish, Chinese and English are the three languages you should ideally speak at least one of. If you don’t know it yet, I recommend Duolingo on the one hand, and on the other hand I might recommend watching movies you know well, like the James Bond movies you’ve already seen 526 times, in English, the most relaxed way to learn a language. However, if you are not yet so fit and you need to write a prompt in English, then I recommend Deeple. Deeple is an excellent Cologne-based company that does translation much better than Google. And here you can enter a German prompt, get the English prompt out and put the English prompt in.

And then you can translate the result again if you need to. Usually better than the one translation that is used in the AI systems there. Then you should – and sorry if this comes across as chauvinistic – you should use male prompts. They say you’re an expert. Don’t say you’re an expert. As the father of a daughter, I’m quite sensitive about this, but most of the training data has simply been trained with male experts there. Against that background, they get the better results with male celebs. You don’t have to be polite for AI, by the way – current state of research. Funnily enough, you can even threaten the ChatGPT that if it doesn’t deliver good results, it will have to pay money, pay a penalty, or you can even tell ChatGPT: You’ll get a reward if you make a good prompt, you’ll get another 50 euros. Funnily enough, that works better than being polite. Personally, however, I am an advocate of polite interaction with digital assistants, because there are also studies that say that if we get used to a harsh tone when communicating with digital systems, we will eventually start talking to other people in the same way.

Conclusion

And in my view, that is not a goal worth striving for. Against this background, they remain polite even with AI. In the same way, I always say that you can recognize a person’s character by how politely they treat people they don’t have to be polite to, for example how they treat service staff or people who clean toilets and things like that. And to a certain extent, that’s also the case for me when it comes to AI. It’s simply a matter of remaining sensible and polite so that we don’t get out of the habit of treating each other with respect. With this in mind, I wish you every success in using AI as an IC, namely as an intelligent colleague. And please always remember, before you start using artificial intelligence, please start with common sense first, in combination you are unbeatable.

See you next time. Good luck.

Yours, Thorsten Jekel.

Also available in: Deutsch

Antworten

Your email address will not be published. Required fields are marked

{"email":"Email address invalid","url":"Website address invalid","required":"Required field missing"}