What can the Chat GPT chatbot with artificial intelligence do or how close is the reign of Skynet?

 

Research and development of artificial intelligence (AI) has been going on for a relatively long time. The tasks that they would like to transfer to it are quite diverse – from template-scenario interactions with objects and people to the calculation of variables and indicators that depend on third factors not currently taken into account by science. These are just a few tasks that are polar in scale, which traditionally can be expanded to infinity according to the imagination of scientists and engineers.

 

In this material we would like to talk about one of the developments that has caused quite a stir, has become available to the masses of people and has quite large capabilities compared to previous developments.

 

What is Chat GPT?

 

Chat GPT (Generative Pre-trained Transformer) is a chatbot that is a combination of a large statistical language model and artificial intelligence. It is stated that he improves himself using guided learning methods.

 

In June 2020, OpenAI announced GPT-3, a language model trained on trillions of words from the Internet. GPT-3 is designed to answer questions in natural language, but it can also translate between languages and coherently produce impromptu text.

 

In November 2022, Chat GPT based on the new model quickly attracted attention. You can chat with him on any topic, and also use him to obtain information on any field of knowledge.

 

 

Moreover, the neural network can generate new products and solutions, for example, write code. You can assign certain tasks to it, and it generates a solution in the form of code. And he can do this in Python, Java, Javascript, and assembler. And now a neural network operating in test mode can be assigned to write basic things.

 

 

Chat GPT speaks many languages, including Ukrainian, so you don’t need to change anything in the settings. The system will detect the language of the request automatically, but it should be noted that automated translation is most likely used. At least, this is the impression received by users who checked the chatbot for the first mentions of specific terms.

 

Thus, Chat GPT in every possible way avoided mentioning that national traitors were first mentioned by Hitler, assuring that the term was started in the USSR. In addition to checking for false data, the chatbot seemed to avoid thorny ethical issues. And the motives for such decisions, such as decoration or completely inventing facts and events, remain unrevealed.

 

The developers themselves warn in the product description that sometimes the chatbot can follow inappropriate instructions and show signs of biased behavior. And also provide, at first glance, plausible, but really false or ridiculous answers. This is partly due to the fact that ChatGPT only owns data until 2022, but beyond this the problems relate to the imperfections of the technology itself.

 

And you can talk about all this with a bot for free?

 

The basic version of ChatGPT is completely free to use. But, of course, OpenAI will not be able to work for free all the time. It is estimated that OpenAI currently spends about $3 million per month to run ChatGPT, or about $100,000 per day.

 

Therefore, OpenAI recently announced a new paid premium version of its chatbot called ChatGPT Plus. You can only join the waiting list. It is already known that the final price will be $20 per month. ChatGPT Plus will provide access even during rush hour, subscribers will receive faster responses and will be the first to take advantage of new features.

 

 

Using ChatGPT is a fairly simple process. The developers allow you to use various tools, such as instant messengers (Slack or FB Messenger) or the Merlin ChatGPT browser plugin, but the main way to launch Chat GPT is to go to webpage.

 

What else can AI products from OpenAI do?

 

Together with the chatbot, OpenAI also developed a service called Whisper, which can transcribe language in multiple languages, as well as perform language translation from these languages into English. According to information provided by the developers, Whisper was trained on 680,000 hours of “bahamian and multi-tasking data” from the Internet.

 

 

DALL·E 2 – another development that can create an image based on a given text description Later, DALL-E 2 learned to draw beyond the boundaries of known paintings using the Outpointing function.

 

What may be the hidden imperfections of AI and neural networks

 

Researchers are studying an interesting phenomenon known as “learning in context,” in which a large language model learns to perform a task after seeing only a few examples, despite the fact that it has not been trained on the task

 

With contextual learning, the model’s parameters are not updated, so it appears that the model is learning a new task without learning anything at all. The researchers’ theoretical results show that these massive neural network models are capable of containing smaller, simpler linear models hidden within them. It was hypothesized that neural network models have smaller machine learning models that can train an older model to perform a new task. To test their theory, the scientists took a neural network model, which is very similar in architecture to GPT – 3, but was built specifically for contextual learning. That is, inside the earlier layers, a model was implemented that trained a linear model, implementing simple learning algorithms.

 

Conclusions and forecasts

 

What to expect from Chat GPT and other artificial intelligence solutions? Yes, you can delegate small projects to bots if you set the task correctly, that is, explain what should be included in the input and what should happen. Often the system does this in different ways and not always fully, but the correctness of the answers is a matter of little time. For now, ChatGPT can be called more of an assistant that speeds up work. To give an example in programming, ChatGPT is already capable of depriving a junior coder of his job.

 

Over time, the neural network will be able to generate basic code directly according to the technical specifications, without many other additional conditions required now. And ChatGPT will not only take jobs away from people, but will also create a new market of specialists to work with it. It is obvious that many truly valuable representatives of professions who will become victims of the neural network will become their trainers. To do this, you just need to master the methodology of transferring experience. For all owners of professions that require systems thinking, empathy, imagination and creativity, that is, a creative approach and properties characteristic of a living person, it is too early to worry.