The flood of articles, reviews, uses (sometimes humorous), anecdotes, fears and discomfort in the face of change seems to be dominating conversations between both friends and technology experts.
I've resisted the urge to post "just another review" of ChatGPT. I wanted to test some of its functions, but above all, I wanted to hear from people who weren't familiar with concepts like Natural Language Processing, Large Language Models or even Machine Translation.
I also wanted to listen to experts and let the dust settle a bit. I have no intention of hiding the fact that this article will be considered provocative: I want to make you think.
ChatGPT is not Artificial Intelligence, it is a Large Language Model.
It just isn't. Artificial Intelligence (more specifically, that "dream" which Artificial General Intelligence was until recently) is the science that uses computers and machines to mimic the reasoning abilities that we humans use intuitively for problem-solving and decision-making.
ChatGPT is a Large Language Model (LLM). That is, a model that has been fed with huge amounts of textual information in many languages and that can predict the next best option extremely well.
It has been trained with A LOT of information: 175 billion parameters and in many different languages. It has also been trained to perform typical disciplines that are part of Natural Language Processing, such as summarization, question-answer systems (chatbots, as we all know them) or Machine Translation systems, for example.
We are not prepared for what is coming
If ChatGPT’s predictive quality, summarization capabilities, results, and the information it provides us with (often from Wikipedia and related sources) already appear to be practically human, it will all seem like a simple toy when OpenAI launches version 4, trained with the 100 trillion (American) parameters and that will be 500 times more powerful in February/March 2024.
ChatGPT has been given a name and not a number for the first time (it is version 3.5), as it is an intermediate version between the powerful GPT-3 and the 2024 GPT-4. Within the language processing field, we would classify it as an excellent text generator, an excellent chatbot, an excellent translator, an excellent summarizer, and many more things.
Honestly, I don't think we're ready for what is coming yet. Sam Altman, CEO of OpenAI considers ChatGPT a toy and an intermediate step that will soon be forgotten. Its code may be released, just as the GPT-2 was released a few years ago. The latter was the basis of various technological projects, such as Maria, which our colleagues at the Barcelona Supercomputing Center created with 135 billion words from the National Library’s web archives, among other sources.
The way it presents information goes beyond the search.
As we will see later, and as the CEO of OpenAI himself makes clear, the company's goal is not to compete in the search market (“What has changed in technology in the last 10 years? Search?”).
We look for process information as we did back in 2012, or 2007. The way ChatGPT presents the information shatters the current model based on results and advertising. It has sent chills down the spines of some defensive "incumbent" tech giants. But its goal and mission are different: to achieve Artificial General Intelligence.
What changes will ChatGPT bring?
Primarily, ChatGPT and language processing technology are going to transform how we access and manage information.
This tool allows us to go much faster compared to our daily interaction with information search engines, which are based on a request (typically a few keywords) and a list of results from which we choose what we think is most relevant or we like the most.
It doesn’t give results. You can't manipulate your way to the first page using loading speed, number of links, or domain authority. The user does not have to digest several results to reach a conclusion. You do not have to reach the second page and take note of the three or four articles or books you have to read to form an opinion on a subject.
Of all the innovations that this "toy" presents, the most revolutionary is the enormous capacity to summarize knowledge. This is a real spanner in the works to the business models of companies that have based their revenue on user searches, trends, etc., but providing merely lists of results and not concrete information on which to base a decision.It transforms reality because changes the way we look for and manage information on a daily basis, which is the foundation of modern economies. Each of us conducts dozens or more searches a day.
ChatGPT offers added value by searching and summarizing information from multiple sources and thus becomes a new tool that allows us to improve our productivity. Best of all, the quality that we currently find mind-blowingly will be old news in one year’s time, with the arrival of GPT-4.
The information retrieval industry is trembling because this tool allows us to manage information in a much more productive way. We can ask it to gather the economic theories of three 19th-century authors, compare them and give us a synthesis - all within a minute or two. That was one of my tests and though the results weren't 100% perfect, it saved me weeks of reading and writing. The result of that comparison and summary was quite decent and I could have written a good article with it, contributing my opinion to add value to my human experience.
That particular experiment also made me realize that among its various flaws is not knowing what is true or false (I insist, it is not Artificial General Intelligence, but a great language model with enormous predictive capabilities). The reflection we have to make is not whether our jobs are in danger, but where this technology is taking us. It is bringing us into a new world, a new reality.
When people start to think about it, fear appears, opportunities arise, and the hype begins. There are those who jump on the "Artificial Intelligence" bandwagon, people who voice opinions who have never had one before, etc.
ChatGPT has become a social phenomenon. And it is these kinds of phenomena that bring about cataclysmic and sometimes revolutionary changes. That is why there are webinars on GPT-3 every day, and that is why we should start recommending to our kids that whatever they do in the future, they start asking themselves: "How can I do my job better with the help of AI?"
Search engines like you.com have already integrated it into their searches (which turn out to be more friendly, though far from perfect in certain places). Microsoft has seen its most direct competitor’s weak spot and, outside its financial plan, has put 10 billion dollars on the table to be part of the OpenAI consortium.
Jobs that may be affected by ChatGPT
All jobs with little or no added value will be the first to be affected. For example, low-value newsletters, especially those that base their information on whether is trending on social media. Article reviewers may also be affected, and even disappear like reviewers of human translation have been disappearing over time (Machine Translation is already of near-human quality on many levels).
From my point of view, any work that depends on systematically collecting information is susceptible to being affected.
A great language model avoids human error, it is far more powerful than any computer, and of course, it is not impressionable. I want to be provocative here, and it might be scary to bring it up, but LLMs can do a better job than some judges.
Because we are living in a two-speed world: one in which we know the technological possibilities that can help us improve, or at least make things better, live a better life; and another where we will have the frustrating reality or realities that we assume are normal when they are not. And this situation of "injustice" is often the cause of revolutions.
There are interested parties who will not want to rethink the situation or for it to evolve or change. But this is just the beginning; frankly, we are not even at the door of Artificial General Intelligence, but we can realize the potential of these systems, even lacking filters and "intelligence". What is genuinely surprising is the speed at which everything is happening and that no one had anticipated.
B) Content creation.
Just like Photoshop or Illustrator have opened up and expanded the artists’ capabilities, content creators will co-create, they won't have to do the work from scratch.
C) Video game creation.
Large parts of video games have replicable common It won't make sense to develop the first 50%, 60% or 70% of the game, and as with content creation, our developers will be able to speed up and improve the processes.
D) Education and learning.
We have a real problem here, I found out today that, in a visionary move, my daughter's school board has banned the use of ChatGPT. We know what prohibitions often lead to, especially when it comes to knowledge and technology. My job as a parent is to encourage my daughter to use every possible tool at her fingertips to process as much information as possible.
If we focus on the relationship between students and this new technology we will realize it isn't as we would expect it to be (something surprising since during the last months we have been told everywhere in media how well they get on). As stated in the 2023 e-Learning Statistics from Studocu, 66% of the students say they haven’t used a tool like this before. Of those who did use an AI tool, 95% indicated this tool was indeed ChatGPT and 36.9% of those who used ChatGPT said they used it to get an explanation on a topic they didn't fully understand. Less than 20% seem to have used it to cheat.
Some studies have already ventured that in the next 5-10 years, half of the American universities may go bankrupt because it is not only the learning but also the correcting that is at stake.
In the 21st century and beyond, we will not need to act as memorizers of facts or links in a production chain. We are going to need our creative side to move forward with the capabilities that AI will augment us with.
The human being, by nature, is not a processor but a creator. The highest job dissatisfaction occurs in processing jobs, not jobs where we exercise our creativity and our cognition. I have recently heard statements such as 'homework is over' and this need not be a bad thing, as long as knowledge acquisition is aimed at our improvement. Correcting homework and exams can also be greatly reduced.
Historical data is not the important bit, but rather the management of that information and what we do with it. The knowledge we receive is meant to work in the system, not to exercise our freedom.
E) Language teaching and the translation industry
Partially related to the above: learning languages and translating content is important, but systems and management add costs that escalate the costs of quality Machine Translation that is often just fine for the purpose.