ChatGPT - Facts and myths for programmers

In just a few months, the number of active users has reached 100 million, which is an astonishing figure. The huge interest is no coincidence, the AI solution developed by OpenAI is one of the largest and most sophisticated artificial intelligence language models available on the market today. In this article, we go through what ChatGPT is, why it's such a big deal, and how it could affect developers of today and tomorrow with Green Fox Academy's CTO, Péter Szél.

‍

What is ChatGPT and why is it such a big deal?

‍

ChatGPT is a chatbot based on artificial intelligence that can provide intelligent answers to user questions as if we were chatting with a real person. We can talk with it about life's big questions, it can generate a literary essay or even a line of code, translate long texts in seconds, and the list could go on. In fact, ChatGPT is a chatbot with a huge language model, which is trained on a large amount of human-generated text. The system developed by OpenAI first went live in November 2022, and reached 100 million active users by February 2023.

‍

"To help you imagine a language model, let’s put it simple: there is a large set of statistical data about the typical order in which words follow each other in a given language. So they collected a lot of data, from which they created a huge database. Thanks to this, ChatGPT can formulate its answers as if we were really talking to a human being," explains our Chief Technology Officer, Peti.

‍

According to Peti, ChatGPT is revolutionary compared to previous chatbots because its predecessors "still had a predefined list of possible answers, and if they detected certain keywords in the question, they would choose the appropriate answer from the list. This new AI technology, on the other hand, creates its own text based on its statistics and database, and doesn’t work with pre-loaded answers."

‍

Even though we are not talking about intelligence in the human sense, this is the closest a computer system has ever come to it.

‍

ChatGPT limitations

‍

The intelligence of this chatbot lies in its contextual awareness on the one hand and its large database on the other. Since the language model is trained on human-generated text, we can say that it isn’t actually its knowledge. If we stop generating new articles, stories or even lines of code, ChatGPT and similar systems will not create anything new on their own. For example, when we chat with the tool or ask to write an essay, “the text itself might be worded in a new way, but at the end of the day the idea behind it isn’t new, so as I said, it cannot invent anything new," adds Peti.

‍

Since it is not able to think, it is more likely to make mistakes, so we cannot trust its answers unconditionally at the moment. Peti has already tried the tool in his own work: "We wanted to set up a live system for a company, and I asked ChatGPT to generate a code that helps to automatically deploy web applications to the server. This is a less exciting task and I thought it might be a good case to test this AI tool. The interesting thing was that at first glance it generated nice code. I copied it right away, but it didn't work. I started looking through and quickly realized the error: it had written keywords into the code that didn't exist, and so the whole thing didn't work. So, sadly, I wrote it myself," says Peti about his user experience.

‍

However, it’s important to note that ChatGPT and similar systems can be a good thought-provoking tool, and can not only speed up certain workflows, but also transform professions, especially those that work with text (e.g. content writing, translation). At the same time, fields that require complex problem-solving skills, creativity and critical thinking, such as software development, are less affected by the use of AI language models.

‍

GitHub Copilot, the programmer's wizard

‍

While ChatGPT is for general use, there are AI solutions that are made for specific domains, such as another development of OpenAI, GitHub Copilot, which is designed specifically for programmers. Similarly to ChatGPT, this tool's purpose is not replacing programmers either, it's more about helping them write code faster and with less work.

‍

"From the large amount of source code available on GitHub, they made a statistic of the order in which the codes typically follow one another, and as the programmer types, Copilot suggests a solution. Similar tools have existed before, like Word or Gmail. These tools will improve over time, and they will give better and better suggestions," explains Peti. He says that these tools can help programmers to work much faster, because they don't have to write each line of code individually, so they can dedicate more time for creative work, and at the end of the day the developers still have to make the final decision if they insert the tool’s code or not.