How to use ChatGPT: a guide for non‑IT people

Google recently updated its privacy policy so that all publicly available user experience will be used for AI training. Social media posts, reviews, blog articles. Until we figure out if it's good or bad. But one thing is clear - artificial intelligence is increasingly woven into everyday life.

Today we will analyze how an ordinary user can take the most out of AI for himself. Let's use the popular generative artificial intelligence ChatGPT as an example.

Read ChatGPT non-IT guide by Briz team and use modern technologies like the pros.

Opportunities and myths of ChatGPT

Unlike YouTube bloggers, ChatGPT is not all-powerful. For example, he can really create an interactive text-based game. If there is enough description and understands what is required of him.

We tried. Even with a minimal description, Chat created an adventure:

All we have indicated is the condition of the game (there will be choices) and the main character is Indiana Jones. That was enough: Chat knows who Indiana Jones is. AI learns from databases, so a well-known hero is clear to him: the choice of setting, the name of the adventure, the format of the story - everything comes from the hero.

But not everything is so perfect.

The chat misunderstood the command. But this is not the fault of AI, but of the operator. After all, it was necessary to write a capital letter D, and not lowercase. This is in the direction of how sensitive ChatGPT is to commands.

Let's try differently. Let's define the game about Indiana Jones again, but use a slightly different query

The details are the same: Indiana Jones again, again the choices marked with letters. But the request is formed differently and ChatGPT created a completely different game:

As you can see, the request format directly affects the Chat response. Even if the preliminary data is almost identical.

Another important point: what ChatGPT knows is limited to 2021. We asked him about Stephen King's latest book. And Chat honestly said that he could not answer, because he only knows information for 2021 and there is a chance that the author wrote a new book:

It becomes clear that the quality of ChatGPT depends only on requests. And now let's talk about how to form them.

How to properly set tasks for ChatGPT?

Commands for artificial intelligence are called "prompts" (prompt). In fact, these are instructions for AI.ChatGPT did a tricky thing: it imitates a real person, so users refer to it the same way. But we must not forget that we have a program that skillfully uses databases. No more, no less. Therefore, it is easier to contact with the help of instructions than to a real person.
 

•  If you refer to what is known for sure and is in the database - give a brief instruction.

Then edit with additional clarifications. For example, the game "Snake" is famous, Chat understands exactly what you mean. And then specify what you need: "change the color of the snake to blue", "make objects more transparent" and so on.

• Be specific if you need to create something new.

ChatGPT can create stories, stories, collections based on its data. But you need to clearly describe what you need.

•  Use English words whenever possible.

Chat does not know how to understand, it picks up data and uses it. Since its databases are predominantly from English-language sources, it works best in English. Therefore, you can insert English words into prompts or write a complete query in English.

• Don't use niche or slang words.

For example, in the gamer community there is a gendarme called "bagel". Each player understands what is at stake, but Chat may perceive it not as a genre, but as a pastry. when you use words with many meanings or surzhik, specify the meaning.

•  Write additional instructions.

ChatGPT knows how to refine the work done. For example, you can ask to continue an already written story. Or, from what he has already written, select the main thoughts and write them down as abstracts. Or rewrite an already created publication, but on behalf of Taras Shevchenko or something. You don't have to download complex instructions right away. Write down one request, and then specify the details you need based on the work done.

ChatGPT has a paid version: it can give more detailed answers, understand broader queries, can count (meaning complex calculations), and many more useful features. At the time of writing, the paid version costs $20 per month. But there are some life hacks that will allow you to extend the free version of the chat.

Here they are:

1.  Answers have limitations. By the number of characters and by the loading of the Chat itself. If suddenly the answer stops, and the “generated” mark disappears, write “Continue” and the chat will continue the answer.
2. Most training data is in English, so query processing in English is more accurate. For example, using jargon, understanding dialect and slang.
3. Ask ChatGP: how to get the desired result with it?

How to live in the world with ChatGPT?

ChatGPT is not omnipotent. This is not a competent expert, he should not be trusted when questions are of a medical, legal or psychological nature. It is only a generative artificial intelligence. This is his goals, tools and tasks - to generate something based on the instructions and the base on which he studied.

✅ Write a story, a book, collect a post on the social network based on the given theses, translate the text - yes.
❌ Write a treatment algorithm, no rehabilitation program for PTSD or depression.

There are more highly specialized AIs (for example, Google's MedPalM 2 for medicine or BloombergGPT for finance). In addition to Chat, there are other artificial intelligences for generating content. For example, Bord from the majestic and terrible Google.

Interesting fact. The UK copyright for a work of art generated by ChatGPT lasts 50 years. And in the US, such works are not considered intellectual property (at the time of writing, but many commissions are still dealing with these issues).

Each AI learns differently, circumvents privacy rights, and exposes public sources. AI developers never talk at all about how their particular technology works. Therefore, users cannot know the full functionality of artificial intelligence.

AI is already capable of more than users think.

Now AI is replacing the paradigms of public space on the Internet. During the forums, anonymity ruled the Internet. With the boom of social networks - public statements and the issuance of a personal opinion in general. And now the point is no longer in the publicity of statements on the Internet, but in where and who will use them.

It remains for us to keep our finger on the pulse and learn to live with new technologies: not to try to stop the inevitable progress, but to tame it. That's why we wrote this guide so that everyone can use ChatGPT like a pro.