How to use ChatGPT: my tips after 7 days of using it daily.

A week ago I have tried ChatGPT for the first time and it has been a game-changer since then. I want to capture the learnings and share some tips on how best to use the program.

On day 1, I set up my account to get started with ChatGPT. It turned out to be very easy:

1. Set up your free account at https://chat.openai.com/chat
2. Write a prompt in a dialogue window (your instructions to AI)
3. Read and review the completion (the output of the model). AI tries to predict what words are most likely to follow from the prompt.

When you’re writing a prompt, you need to think about the kind of completion that you want and specify that clearly to the model. What is happening behind the scenes is that your instructions will help ChatGPT select the best possible output. A good way of looking at this is that ChatGPT has basically read the entire internet, so it knows how to use language in a billion ways. Let's see how ChatGPT describes itself:

On day 2, I found out that the program speaks multiple languages. I asked some offbeat questions in English, German, Russian, and even in Turkish to explain my frustration to the taxi driver for overcharging me. Interestingly, AI also has solved random riddles in multiple languages.

On day 3, I decided to be specific with questions and set tonality to them. Asking more precise questions has resulted in better answers. Let’s suppose I want to explain what AI is to a 5-year-old and play with prompts on how best to explain it. Here is a very generic prompt:

The 5-year-old will not understand such an explanation. Let’s try to make it better by giving more context and clearer instructions. The art of prompt is in constructing clear commands and playing around with them.

The results are quite different, right? I was fascinated and decided to start googling frameworks and best practices on writing prompts.

On day 4, I found a framework called CIDI, which in my eyes is the easiest to use:

• Context: a simple sentence to help ChatGPT understand the context you want it to be. A simple way is to write “Act as a …”.
• Instruction: what you want to get out of ChatGPT.
• Details: optional, here you can specify the tone of voice or formatting you want.
• Input: the input data you want ChatGPT to respond to, this can be included in the instruction too if it’s short.

Now let’s try to play with it and get some ideas for a new marketing strategy for your new product based on CIDI framework. Here’s a simple prompt, that creates an average answer:

From above you can feel, that answer is not providing a meaningful strategy. Let's rewrite the prompt using CIDI framework:

When I look at the result that was produced within 10 seconds by AI, I can't believe how complete it sounds. It would have taken at least 1 hour to put such a detailed plan together for someone who is not regularly in touch with the topic. Now let's review the prompt structure and how CIDI framework worked:

• Context: You are a marketing executive at a large consumer goods company.
• Instruction: Write a marketing strategy for the launch of [INPUT]. Include a strategy to understand the right pricing.
• Details: Include a strategy to understand the right pricing. Use a bullet point list format and a concise, formal tone.
• Input: Use a bullet point list format and a concise, formal tone.

On day 5, I discovered the database function. When you ask a question, ChatGPT searches through its database to find the best possible answer. As you see from the outcome, the AI is not yet 100% correct (mistakenly taking London as the European Union capital), thus it is important to double-check the answers.

On day 6, I tried asking follow-up questions and seeing if AI can correct itself. And it has worked! Apparently you can also ask your answers to be shorter/ longer/ more detailed/ written in a different style.

On day 7, I used ChatGPT for inspiration. I'm currently writing a book and was curious if I can gain some of the alternative prospects on the content. Interestingly, I was provided with a new perspective that I did not even consider. Additionally, AI barely makes any spelling or grammar mistakes, so for someone who is writing as a non-native speaker - it helps to sound clearer.

Wrapping up ChatGPT and all the Large Language models (there are plenty of such programs available online) are fairly new technologies that are evolving fast. It’s up to us to experiment with them, test their capabilities and limitations, and decide for ourselves how we want to integrate them into our workflow.

I encourage you to stay curious and follow this space closely. A lot of interesting things will happen in the next few months, and staying ahead of the curve will allow you to have a fruitful and rewarding career!

Very truly yours,

Ilona B. Schukina