How to use ChatGPT correctly?

I know what you're thinking. Another article about artificial intelligence and ChatGPT. Počasi ste jih že naveličani, saj se skoraj vsak dan srečate s kakšnim člankom o tem, kako je umetna inteligenca poskrbela za nov mejnik, škandal, odpuščanja ali kaj podobnega. Potrebno se bo navaditi. To je najnovejša tehnologija, ki ne bo zlahka izgubila zagona. Prej bo dobila dodaten veter v jadra. Glede na to, da smo bili na sejmu MWC 2024 skoraj na vsakem koraku bombardirani z besedo AI, je jasno, da smo »obsojeni« na umetno inteligenco – vse dokler se ne pojavi novejša tehnologija.

ChatGPT is the father of this technological craze. Very soon, close relatives also appeared - Copilot, Google Gemini (formerly Bard), Claude, Jasper... - but they failed to dethrone the originator of everything. Regardless of the mixed feelings you may have about such language models, it is undoubtedly one of the most disruptive technologies of this millennium. It has changed the rules of the game and therefore it is important that we also use the new technology correctly.

Prepogosto pa naletimo na uporabnike, ki ChatGPT in tudi ostale klepetalne »bote« uporabljajo preveč enostavno. Zagotovo ste to opazili tudi sami. Mešanje različnih tem v enem besedilnem pozivu, pa pozivi, ki so preveč splošni oziroma premalo natančni, pozivi brez določenega konteksta in podobno. Ali najhuje – enostavčni pozivi, pri katerih je ChatGPT prisiljen ustvariti vsebino iz ničesar. V takšnih primerih se zelo hitro pojavijo floskule, nepreverjena dejstva in na splošno besedilo brez substance.

Using the ChatGPT bot doesn't have to be complex. With just a few tweaks, you can get much better results. Here's how to do it.

What is GPT-4 (Turbo)?

Preden “se vržemo” v raziskovanje, kako pravilno uporabljati ChatGPT, nekaj besed o jezikovnem modelu, ki deluje v ozadju. GPT-4 Turbo je najnovejši veliki jezikovni model podjetja OpenAI. Trenutno je na voljo samo za razvijalce. Ko bo na voljo množici, bo zagotovo del plačljive različice ChatGPT Plus. Tudi Microsoft v Copilotu uporablja model podjetja OpenAI, Google he has his own (Gemini).

Since many of us are not developers, GPT-4 is the latest model we have access to. Simply put, it is an upgrade from the previous one. It can extract data until April 2023 (GPT 3.5 until January 2022). He is also more reliable, creative and understands more in-depth instructions. It can accept textual as well as image inputs, based on which it can output text similar to human work. DALL-E, a photo generator that was previously separate from the ChatGPT bot, is now available in the ChatGPT Plus subscription.

ChatGPT Plus contains the latest GPT-4 model and also DALL-E

How to write good prompts for ChatGPT?

Did you know that there are professions that deal with creating prompts for language models? And for this, employers pay up to 300,000 $ per year. If you might be thinking about changing careers now, it's a good idea to start with the basics.

Treat ChatGPT as a colleague or friend

Too many users talk to ChatGPT like a robot, too monotonous and impersonal. Imagine a scenario where you talk to a bot like you would a colleague in a meeting or a friend over coffee. And now think about how many times you had to ask a friend or colleague a different question or explain an idea further so that they understood the point. Use the same principle with the ChatGPT bot.

With the difference that your AI friend doesn't need a name, unless you're afraid that AI will one day stage an uprising like Skynet in Terminator. If this is the case, I recommend that you also thank him after each answer.

Just like you lead a friend in a conversation, you lead ChatGPT. If you notice that he's gone off the rails, stop him and review your original call and possibly improve it with additional information. You can start with a simple question or prompt. When you get the answer, ask him another question and then continue in the same rhythm until you get the desired result.

Simply put, talk to ChatGPT.

Be specific and give it context

You can use simple prompts for general answers.

• When was France Prešeren born?
• How would you translate "raytracing" into Slovenian?

However, if you want more specific content, it pays to add context to your text prompt.

Instead of "Prepare me a schedule for traveling around Slovenia" use "My girlfriend and I want to travel around Slovenia". We have 10 days. I am a lover of art and architecture, she feels better in nature and loves natural sights. I would travel by car, so transportation is not an obstacle. Can you suggest what and how we could see as much of Slovenia as possible in 10 days?"

It will take you a few seconds more, but you will get a much more informed answer. It doesn't hurt to limit it from the start. For example, set him a 300 word limit so he doesn't ramble on indefinitely.

ChatGPT works better if you give it a title or identity

ChatGPT is, if nothing else, very flexible. He can speak like Shakespeare or in the style of a teenager with all the slang terms and street talk. If you use it at work, give it a title that is appropriate for the given task. If you want him to write a marketing strategy for a certain product launch, at the beginning, tell him to put himself in the role of a marketing director or something. If you would like him to describe certain products, he should be a professional writer, journalist, editor...

At home, when you're browsing the web and perhaps find a product you want to buy, you and the ChatGPT bot can rate the product together. Put yourself in the role of a consumer and list the disadvantages and advantages as a normal user would notice them.

That said, you can also assign a style and register to it. He should write like your favorite author, you can tell him to make the text relaxed, strictly professional, salesy, educational and so on. Also tell him to take into account that the answer is intended for a specific target group (business partner, public, children...)

Guide him through the conversation

ChatGPT can lose its thread very quickly. If you notice that he has gone astray while he is writing, stop him and add to the prompt that he should only stick to the given topic. When you get an answer, you can always ask him what his answer is based on or where he got his information from.

You'll notice that he likes to repeat introductions, endings, and intertitles a lot. And he especially loves compositions like: "in the coming..., in this digital environment...". If you don't want the text to come off as generic, either fix it manually or warn ChatGPT from the start that it shouldn't use it.

Don't be afraid to experiment

If you think your prompt is good, there is no need to change it. You can just repeat it and ChatGPT will generate a completely different response. You can also tell it that you want data in the form of a table, or that it extracts the most important information from a certain text/article in a few paragraphs. You can use it in a hundred and one ways. You can sort the information alphabetically or by some other criteria. You can also cheat and tell it to come up with the best text prompt on its own to come up with the text or photo you want.

As long as you keep the conversation open, ChatGPT will remember your conversation and your preferences. It's not 100 % reliable, so it's a good idea to do multiple conversations for different scenarios.

You can also create your own GPT chatbot, more on that next time.