What is ChatGPT; the chatbot that turned the internet upside down?

From Twitter to tech media and Microsoft, ChatGPT is everywhere these days; what's the reason for all the fuss about the mysterious company OpenAI's chatbot?

مرکز رؤیت هلال امارت اسلامی افغانستان به همه هیئت‌های ولایتی و ولسوالی‌ها اطلاع داد که امروز، سه‌شنبه، پس از ادا نماز عصر، برای رویت ماه مبارک رمضان در مکان‌هایی با بیشترین امکان رؤیت حضور یابند و با دقت کامل تلاش نمایند.

مسؤولین موظف شده‌اند تا یافته‌های دقیق خود را از طریق محاکم ولایتی به هیئت مرکزی رؤیت هلال گزارش دهند.

این اقدام با هدف اطمینان از رویت صحیح ماه رمضان و هماهنگی دقیق میان هیئت‌های ولایتی و مرکز انجام می‌شود.

Until recently, AI chatbots were pretty awful. But ChatGPT is different from all the chatbots that came before it. It’s smarter, weirder, and more flexible. It can tell jokes, code, and write academic papers. It can even diagnose diseases, create text-based games set in the Harry Potter universe, and explain the most complex scientific topics in the simplest way possible. ChatGPT has caused such a stir on Twitter and the tech media that almost everyone is talking about it these days. Some have gone so far as to call it a replacement for Google’s search engine, while others say its impact on the world will be on par with the iPhone. Meanwhile, there are those who say that ChatGPT has stored all of humanity's scientific knowledge and is close to reaching human-like artificial intelligence and a level of self-awareness. Microsoft also plans to invest $10 billion in the company that created this chatbot and bring this technology to the Office suite and even the Bing search engine before the end of March. The only chatbot that successfully passed the Turing test and fooled users into believing it was a computer was ELIZA, a software program designed in 1966 by German-American computer scientist Joseph Weizenbaum that responded to users’ responses by matching simple keywords. ELIZA was essentially a simulation of a psychologist who would ask the user to describe their problem, then search for the keyword for the appropriate answer, usually continuing the conversation with another question. Now!, 57 years after ELIZA was launched, chatbots have reached a point where the world is left speechless with their amazing and miraculous abilities. And among them, ChatGPT is the best chatbot that the world has had the opportunity to use.

ChatGPT is free to use; but unfortunately, accessing its website is almost impossible for us in Iran. Unless you use an IP change tool and have a real phone number from another country, for example, the United States (virtual numbers do not work.) If this is not possible for you, you can try your luck with this chatbot’s Twitter bot. Many Twitter users ask ChatGPT a question they want it to answer by tagging it @ChatGPTBot, and if the bot is honored and answers their question, it retweets it on its page with the answer. (Although I should also say that this bot has not yet answered any of the questions I have asked it, and you will have to be either very patient or very persistent!) In this article I have tried to cover all the dark and bright sides of ChatGPT. I hope you will follow me until the end of the article. What is ChatGPT? ChatGPT is an experimental chatbot, or rather the best chatbot that has been made available to the public so far. It was created by OpenAI and is based on version 3.2 of the GPT language model developed by the same company. OpenAI is the mysterious company that developed the highly intelligent Dall-E image generator and wants to be the first to achieve human-like artificial intelligence. GPT stands for Generative Pre-trained Transformer, an example of artificial intelligence technology known as a large language model that is automatically trained over a period of weeks using an incredibly large amount of data and computing power, and generates text based on what it has seen and learned during this time. OpenAI initially planned to make the source code of all its projects freely available to the public; however, after the open-source release of the second version of GPT in November 2019, it permanently abandoned its non-profit structure and blocked public access to the source code of its future projects, including updated versions of GPT. Although GPT-2 was a significant achievement in its time, due to its limited dataset and only 1.5 billion parameters (a pattern in language that Transformer-based models use to predict meaningful text; similar to the connections between neurons in the human brain), it lacked the power to produce text that wouldn’t become repetitive and boring after a few paragraphs. The third version of the model was released in beta in June 2020 with 175 billion parameters, making it more powerful than the LaMDA AI with 137 billion parameters. Interestingly, Microsoft has exclusive access to the GPT-3 source code, and ChatGPT and the GPT 3.5 language model were trained using Microsoft’s Azure AI supercomputing infrastructure. GPT-3 is a machine learning marvel. The quality of the examples produced by this AI is so amazing that it’s hard to tell they weren’t written by a human. And now, an updated and much more advanced version of this model has been used in the ChatGPT chatbot, which finished training in early 2022 and was made available to the public for free in late November. Much has been said about the wonders of ChatGPT. Users type their requests into the chatbot’s extremely simple interface and receive amazing results; from producing poems and songs and screenplays to writing articles and code and answering any question you can think of; all in less than ten seconds. It would take a thousand human years to read all the data contained in ChatGPT ChatGPT can handle even the most bizarre requests; for example, it can piece together a story where seemingly unrelated details in the first paragraph are connected to events in the last paragraph (something that rarely happens even in the most popular Netflix series!). It can tell a joke and even explain why it’s funny. It can write catchy leads in the style of various magazines and authors, and even use quotes that seem authentic but are completely fake. All of these capabilities make ChatGPT a highly entertaining and addictive tool, and of course, for people like me who write for a living, it can be very disturbing and troublesome. The amount of data that ChatGPT was trained on is so vast that it would take “a thousand human years” to read it all, according to Michael Wooldridge, director of fundamental research in artificial intelligence at the Alan Turing Institute in London. He says the data hidden inside the system holds an infinite amount of knowledge about the world we live in. How does ChatGPT work? ChatGPT is based on an updated version of GPT-3, a large language model (LLM) that relies on a huge network of artificial neurons that, in a way, mimic the behavior of neurons in the human brain. The GPT language model is based on Google’s Transformer neural network architecture, which Google has used to build its advanced LaMDA language model; The same model that a few months ago, a Google employee claimed to be “self-aware,” sparking a wave of internet jokes and concerns about the emergence of killer bots. The Best AI Chatbots The Best ChatGPT Alternatives from Perplexity to Poe By Google’s own definition, “A Transformer builds a model that can be trained to read a large number of words (say, a sentence or paragraph), pay attention to how those words relate to each other, and then predict what it thinks the next words will be.” In other words, instead of analyzing the input text step by step, a Transformer examines entire sentences at once and can model the relationships between them to better understand the content of the words and the context in which they are used. Since this network performs the analysis in its entirety and all at once, it requires fewer steps, and in the field of machine learning, the fewer the number of data processing steps, the better the result.

ChatGPT’s own explanation of how it works In general, AI models of large language are “fed” hundreds of billions of words in the form of books, conversations, web pages, and even Twitter and other social media posts, and the AI ​​uses these vast sources of data to build a model based on statistical probability; that is, the words and sentences that are most likely to come after the previous text. In this way, language models are a bit like the word prediction capabilities of smartphones, except that they operate on a much larger scale and, instead of predicting a single word, can produce complete answers consisting of several paragraphs. The method used to train the language model used in ChatGPT was to first give it a large number of questions and answers that had been handpicked by experts in the field. Then, these questions and answers were included in the model’s dataset. In the next step, the system was asked to provide several different answers to a very large set of diverse questions, and human experts ranked each of them from best to worst. ChatGPT’s mechanism is similar to the word prediction capability on a phone, but on a much larger scale The advantage of this scoring system is that ChatGPT can, in most cases, surprisingly identify what answer each question seeks and, after collecting the right information, how to present it in a natural way. For example, if you ask this chatbot a vague question that it cannot answer precisely, it can correctly guess which subject category your question falls into, such as scientific, philosophical, political, etc., and give you an answer appropriate to this category. Another important thing to know about language models like GPT is that they are a type of generative artificial intelligence (Generative AI). These systems use data previously generated by humans to produce new results using a technique called machine learning, which appears to be unique, but is actually nothing more than a new arrangement of previously generated data. From this perspective, generative AI does exactly what we humans do for various tasks; that is, we use what we have learned from others to develop our own unique skills; except that the learning speed of AI is infinitely faster than that of humans, which is what makes ChatGPT so magical. In fact, this chatbot has processed more text than any human in its entire life, which has allowed it to answer any question better and more fluently than you and me, and even imitate certain people, making its answering style similar to theirs. Another reason that makes ChatGPT the best chatbot in the world is that most AI chatbots are “stateless”; meaning they are unable to store data generated in previous conversations and therefore, cannot remember or learn anything; but ChatGPT can remember what the user has previously told it and convey the feeling that you are really talking to an intelligent robot. The goal of developing generative AI is to save users from doing boring, repetitive and time-consuming tasks by imitating human behavior; such as tools that automatically edit images, correct typos and, like voice assistants Siri and Google Assistant, speed up many smartphone-related tasks. But just as it takes specific skills to search the Internet, using ChatGPT effectively also requires skills to articulate exactly what you are looking for; If you can’t explain your request in detail, the chatbot won’t be able to provide the exact answer you’re looking for. ChatGPT Uses; What’s the big deal about ChatGPT? The technology behind ChatGPT is nothing new. It’s built on the GPT-3.5 language model, which is an updated version of GPT-3, and was released to a limited number of users in 2020, causing quite a stir. A Twitter user asked GPT-3 to examine a laptop possessed by Richard Nixon, in the style of tech YouTubers. He was surprised by the result, tweeting: “Human YouTubers, God bless you. You can never compete with content like this.” So what’s all the fuss about ChatGPT, and how did it manage to reach a million users in just five days, according to OpenAI CEO Sam Altman? (For comparison, it took Facebook and Instagram two years to reach the 1 million mark.) While large-scale language models are not new to AI researchers, this is the first time such a powerful tool has been made available to the public for free, with a remarkably simple web interface. If you recall, the original and much more powerful Dall-E 2 didn’t become as popular as the Dall-E Mini knockoff, because it was only available to a select few, while anyone could easily use the Dall-E Mini (which was renamed Craiyon due to copyright issues).