The problem is that we often confuse learning systems/neural networks with artificial intelligence. These are completely different concepts and systems. And the goals are different.
ChatGPT can hardly be called an artificial intelligence or a "matrix"
In fact, it is a huge repository of information that can find the right answers among trillions of blocks of information and formulate answers similar to human speech. A sort of Google search engine is a wrapped human-like interface for issuing search results.
Yes, neural networks and other technologies work here to collect and systematize primary information. But this system has no INTELLIGENCE.
I hope you have already seen this masterpiece: "The neural network was asked to draw a salmon swimming against the current. Well, she drew it."
In ChatGPT, the picture is much better, but the intelligence never appeared there. Since one of the areas I do is IT (development), and like everyone else I saw how ChatGPT "produces ready-made ideal code." I tried to set a task for him, something from the area "Write code so that when you click something, something opens somewhere." The system began to ask clarifying questions so that the problem statement was extremely accurately described in order to understand which "code from the knowledge base to take." If there is no clear statement of the problem, you will not receive any code. More precisely, if there is a place where there can be variability and you need to figure out the essence of the process yourself, ChatGPT will not help you. Therefore, if we talk about "replacement" - perhaps he will replace some of the "pioneers" of developers, who are essentially "translators" from human to programming language, not difficult tasks. Although I agree that as the system learns and the knowledge base accumulates, the level of "replacement" may rise
You do not seem to understand what ChatGPT is and confuse it with generative-adversarial neural networks for generating pictures, such as Midjourney or Stable Diffusion.
Well, actually, I participated in the process of testing this system, I have been using it for my own purposes for a long time.
Explain what you wanted to invent, but, as always, did not express it very skillfully?
Surely you can tell how he can write lyrics, draw pictures and write music on his own? It's not Artificial Intelligence, don't worry
Or did you mean something else? Then explain? Just not the classic answer of a person who read the title somewhere, he doesn’t understand, and with any more or less complex question he answers something like “Well, what will I explain to you if you don’t know anything about him” or “search the Internet, I don’t have time to explain such simple things"
If you start to study the subject area and not snatch out "beautifully sounding abbreviations", then it is possible that you will understand what it is about, and what, for example, Neural Networks, Deep Learning, Machine Learning, Artificial Intelligence, decision-making systems, and much more. And most importantly - the DIFFERENCE between these concepts
UPD
Oh, I think I understand - did you decide that this example shows the work and essence of ChatGPT ?
)))) no, I showed that such powerful systems, almost endowed with "intelligence", are just processing and projection of those primary data with which they were trained. And I will add - in this example, the system knows exactly what the salmon fish looks like and not its fillet, but it cannot really understand the essence of the issue and compare the need, and draws the most frequently downloaded and requested "tag" with other entities. And only by adding this "error" to the training dataset, the system will once ask - "can you draw a whole live fish or steak"
But by the way, the formation of answers and questions is a different technology, and it is called LLM - Large Language Model. Plus a deep learning system.
Well, about texts and pictures - this is all the result of the work of the deep learning system, Dall-E 2, this is an "engine" that works separately from ChatGPT, but supplies "content for answers"
In the GPT-3 version, the system was trained on 570 gigabytes of initial "knowledge". On the basis of this, about 200 billion parameters were formed. I can tell you a lot more, but I won’t waste time, you have the American Internet at your fingertips, the American ChatGPT - they will help you
Although I agree that ChatGPT is called "artificial intelligence", and the name of the company unambiguously flaunts "Open + AI" or open artificial intelligence.