ChatGPT: Beyond the Hype
With all the ongoing buzz around AI and ChatGPT, many non-technical users seem lost in a mythic space of exaggerated facts and arbitrary speculations, with emotions swaying between horror and religious frenzy. This post aims to bring clarity and common sense into this space, helping readers take a more balanced view of this new phenomenon.
As a developer, I became interested in integrating AI tools into my web projects a few years ago, which recently led me to build a few experimental AI-based web apps like this one.
ChatGPT is an outstanding product, and I believe that’s why it has garnered so much attention. It consistently delivers excellent results, particularly in conversation and writing, supporting the writer throughout the entire process, from the initial outline to the conclusion. Its manner of speech is remarkably human-like, resembling a knowledgeable, welcoming, supportive, high-profile consultant. Moreover, unlike many specialized projects, it is a general-purpose AI—an experiment paving the path for AGI (Artificial General Intelligence).
The technical side
ChatGPT is a chatbot developed by the OpenAI team. The underlying OpenAI project was supported by influential figures such as Elon Musk (who later withdrew) and partnered with Microsoft gradually increasing their investments it from 1B to 10B to 100B USD—an amount comparable to the yearly budget of a mid-sized country. Other key players, like Google, have similar projects, but they appear to be following in ChatGPT’s footsteps.
Under the hood, ChatGPT utilizes a large language model using GPT (Generative Pre-trained Transformer), a specific type of neural network as the underlying architecture enabling the language model to understand and generate human-like text.
I won’t delve into the details of how a neural network works, as it’s a complex and time-consuming topic. Suffice it to say that it’s a mechanism for statistically evaluating the relationship between input and output data. When the output closely matches the expected result, we can say that we have a functioning AI.
Nothing about this is extraordinary, except for the large numbers involved. For instance, GPT-3 boasts 175 billion parameters (to put it into perspective, the human brain has approximately 86 billion neurons), and there are suggestions that GPT-4 may have one trillion weights (although the direct correlation between the number of weights and AI power is not certain). Additionally, GPT-3 was trained on vast datasets of Internet text, totaling 570GB (equivalent to 1.4 million average-sized books).
Thanks to these numbers and the team of world-class AI experts, GPT produces awe-inspiring results. Given a sequence of words (or “tokens,” i.e., normalized words), GPT can predict the next word with high probability. This ability enables GPT to perform cool tasks like answering questions in a human-like manner, generating text or images, and providing various other services.
The current state of things
Efforts to create something resembling human intelligence have been ongoing for a long time. However, the true breakthrough came only a few years ago when the immense potential of large language models was discovered.
“I knew I had just witnessed the most important technological advance since the graphical user interface,” said Bill Gates, describing his pivotal meeting with the OpenAI team last year. — “The development of AI is as fundamental as the creation of the microprocessor, the personal computer, the Internet, and the mobile phone. It will change the way people work, learn, travel, receive healthcare, and communicate with each other. Entire industries will revolve around it. Businesses will distinguish themselves by their adept use of AI.”
However, at present, we are still primarily dealing with the demonstrated ability to maintain a human-like conversation. While this is a significant achievement, it doesn’t equate to human-level intelligence. In its current state, ChatGPT is, in many aspects, simply a novel format for presenting information, creating an impression of dialogue.
How can an ordinary user benefit from AI right now?
Currently, the popular “There’s AI for that” project counts 4,257 AI projects providing services for 1,188 tasks. Most of these tasks involve:
- Generating texts in various genres
- Generating images, videos, audio, presentations, and other media content
- Answering questions
- Facilitating learning and research
- Assisting with business tasks such as customer support, lead analysis, advertising, security, documentation, etc.
In this extensive list, almost everyone can find ways to personally benefit from using AI.
Of course, we can still live without these fascinating tools, just as we lived without mobile phones in 1990, without the internet in 1996, or without e-commerce in 2000. There’s no need to rush out of fear of being outperformed by those actively using AI. Sooner or later, we will find ourselves in the midst of an AI-driven world, perfectly adapted without much thought. The ones rushing are those who wonder how to make use of this new reality before it becomes commonplace—a strange mix of heroes and frauds, saints and sinners, driven by both the loftiest ideals and basest greed, shaping history.