ChatGPT: with us or against us?

— Spoiler: ChatGPT will not replace developers, business analysts, and testers in the near future. It will not even leave journalists and writers without work. Yes, this is the first AI with the rudiments of creativity, but it is not able to visualize, often comes to strange conclusions, has no life experience, and lacks soft skills. So far, it is an upgrade of the search engine, but with huge development prospects.

It is quite possible that ChatGPT will officially become the first chatbot to overcome the Turing test because it perfectly supports dialogue and knows how to lie. Although I bet that artificial intelligence will fail the test intentionally. Google Translate has not yet replaced translators, Alexa has not left teachers out of work, Siri has not yet replaced anyone’s friends, and the chatbot "Friend," invented in Ukraine, has not become a professional psychologist. Show a like if you also clean up after your robot vacuum cleaner.

You can argue that there are two different levels of comparison, and that it is more appropriate to compare GPT and Google Assistant, but these are just conventions to outline the problem. And, as you may know, Google has already offered limited access to a chatbot technology with artificial intelligence called Bard to compete with GPT.

Should we be afraid or not?

— Stephen Hawking wrote that artificial intelligence could end the era of humans. But, judging by what I see, now it can only put an end to the era of copywriters and content creators. Perhaps it will give us a new profession — "ChatGPT operator" (Prompt Engineers already exist), who will formulate questions as effectively as possible and check the information received in response.

Now, GPT is able to add a qualitative and plausible (at first glance) continuation of any prefix text, but it is a generative, not an analytical, tool. It is capable of minimal training within a single chat/request, which means that it can be programmed. But you need to understand that it creates answers. This is not a source that can be referenced, it is fully generated text that in many cases only appears to be adequate. Facts may be intertwined with complete fiction in one sentence. For example, when I asked "Who am I?" ChatGPT replied that I was "a world-famous designer of Ukrainian origin."

What can ChatGPT do?

— GPT is already a powerful tool for solving daily problems: it creates feedback and letters in correct business language, can help create a presentation, can formulate requirements for a vacancy, and it can be applied very successfully in the field of education. It recently passed exams in a number of disciplines (business, law, and medicine), qualified as a third-level coding engineer at Google, surpassed most microbiology students, and almost passed a literature test. It is obvious, however, that so far it lacks the critical thinking of an ordinary person.

Who is ChatGPT useful for?

Business Analysts

— For business analysts, its value lies in the fact that, in a very short time, it can: look for inconsistencies in requirements; generate brainstorming questions (Can you please give me 10 questions I could ask a customer service representative about this feature? [Feature description]); help describe project documentation; plan an event; and similar operations. ChatGPT can offer insights to help understand market trends, and possible customer behavior. It does an excellent job with use cases, and can adequately break them into user stories, which, of course, will need review. But, even if they are limited, the things it can do are already a starting point for you; a ready-made draft that does not need to be created from scratch, a list of issues that you do not need to think about for a long time.

Data Analysts

— GPT can help you explore data, find patterns, and even create relevant visualizations. Using human language, it can explain complex concepts and ideas in easy-to-understand terms.

Developers

— It also provides great prospects for developers — answering questions like "What libraries can I use to solve the problem?" before generating the functional code. I also recommend using GPT for debugging during training, so that it explains what a particular error means, along with StackOverflow.

It should be understood that ChatGPT can help with ideas, but not with implementation. Only you know what customers need, what kind of personalities they have, and exactly what can bring the most profit to their company.

Areas of application and problems

— It seems that this technology can also become an indispensable tool for Customer Support. GPT can be used for:

• round-the-clock customer support;
• the entertainment industry (individual recommendations for movies, music, etc.);
• optimization of legal services (lawyers can use GPT to provide fast and effective legal advice to their clients);
• recruiting and staffing (automation of the recruitment process and assistance in interviews, which saves time and resources);
• sales and marketing areas; and
• travel (booking tickets, selecting routes), and more.

A separate point here is medicine. MedPaLM, medical GPT, is also developing on the horizon. This literally means the apocalypse for hypochondriacs, because Googling symptoms will now become even scarier.

A similar dilemma is posed by the most discussed case of the bot passing the exam in astrophysics. How will artificial intelligence affect the education system? How exactly will it help to develop it?

AI tools create completely new problems for everyone: from associations to regulatory bodies, from universities to private companies. Not surprisingly, there is considerable confusion around AI solutions, their potential use cases, and the legal/ethical/technical implications of their implementation.

What can ChatGPT not do?

— ChatGPT very effectively sums up and paraphrases other people's words and, based on what we create, can offer several well-known examples, change the composition of the text, translate it into another language (enough to understand, but still not as good as DeepL). The quality of the information received from ChatGPT, however, depends on who formulates the question and what material was provided to it. You should not expect ChatGPT to write a good scientific article if you do not give it materials. The output will be generic empty phrases. The person who asks it a request is fully capable of controlling the resulting output. If you leave the result to be created exclusively by artificial intelligence algorithms, then there is no hope for a high quality outcome.

Surprisingly, ChatGPT does not solve Olympiad math problems well and, for those, everything you need is out-of-the-box thinking. It is most difficult, however, for ChatGPT to solve dilemmas that a human intellect understands immediately.

Managers definitely don't have to be nervous. ChatGPT will also not be able to pass the Caliper test, which evaluates mental and personal data and can predict how a person will work in a certain position, for many more years.

Fortunately, so far our brain learns better than a machine. The human brain is more plastic and much more complex than AI; it draws conclusions from previous experience and connects them with other knowledge. Artificial intelligence will also soon be able to do this, but will it completely replace real intelligence? In my personal opinion, that would be a matter of more than one decade. Despite technological breakthroughs, the capabilities of modern computers are limited. Typical deep learning algorithms simulate only a small part of the function of the human brain, which corresponds to the initial stages of information processing. In our brain, this happens unconsciously and takes up to three hundred milliseconds, while machine learning requires billions of training attempts.

Tips for using GPT with maximum benefit

• Use English for better interaction;
• The chat database only includes events up to 2021, so don't be surprised if it gives outdated information;
• Your request must be specific (specify the programming language, input data, output data format, etc.);
• If you are just learning to program, the chat can help you find errors in the code or offer your own solution to the problem;
• Chat can prompt a solution, but will not do the work for you;
• Try sending the same request several times, the answers are always different;
• Combine the official ChatGPT and Google Chrome Extension Merlin;
• Don't use profanity, it's programmed to ignore it;
• Always provide context or clear, specific points to your request if you ask it to create an essay or article. Don’t forget to specify that the text must be unique, otherwise it will not pass the plagiarism check;
• Write correctly, GPT sensitivity to grammatical and spelling errors is currently limited;
• Always check everything you receive as a result (many of the links provided by the bot don't work); and
• Remember that you are the authority in establishing the task, so do not immediately believe everything that the chat offers.

Today we are all free ChatGPT mentors. The more questions we ask it, the better it will work in the future.

Our thanks to our partner site wearecommunity.io.