The American company, Open IA, a pioneer in the United States in the development of generative artificial intelligence (generative AI), reached its peak with the publication on November 30, 2022 of its conversational agent, Chat GPT. The latter is considered a revolution in the field of generative AI. Because it manages to answer increasingly complex questions and with more coherence. It is now more difficult to differentiate between text written by GPT chat and a human. This is explained by the colossal amount of data on which it was trained.
PhonlamaiPhoto/Getty Images
Since its birth, artificial intelligence has never missed an opportunity to provoke diverse reactions. Between myth and reality, today generative artificial intelligence (breaking latest news) generates all possible and imaginable emotions. Indeed, since November 30, 2022 with the launch of Chat GPT by OpenAI; American company initially non-profit, now transformed into a capped profit company; this generative AI has ignited the web both with its ingenuity and the curiosity it arouses. But contrary to what many Chat GPT users would imply, generative AI is not new; and Chat GPT, is also not the first in this category of AI developed by the company, led by Sam Altman.
What about generative artificial intelligence?
The Dall-E 2 tool home page: screenshot of the Express
As a reminder, artificial intelligence is defined as the ability for machines, more precisely for computer programs, to imitate human intelligence. This definition appears to be very limited, as artificial intelligence originally focused on a single task only. For example, we could program an AI to cook, play the role of receptionist in a company, receive orders, etc.
Read also: Artificial Intelligence: its birth and evolution.
But today, artificial intelligence, particularly generative AI, is part of a much more global perspective. Thus, we define generative artificial intelligence as being the capacity for a machine or a program to reproduce human cognitive capacity in a global or versatile manner. That is to say, generative AI is not limited only to a specific and simple domain, sector or activity. On the contrary, it can perform diverse and varied tasks. For example, it can produce coherent texts and answer questions relating to different disciplines, such is the case with Chat GPT.
Other existing AI of the same type can generate images that are barely believable, taking into account their realistic nature. This is the case with Midjouney, one of the most popular generative AIs. But there are other generative AIs besides those mentioned here. We think, for example, of Dall-E, also developed by Open IA.
The latter is a generative AI that can generate very realistic images from simple descriptions made by users. This generative AI is equipped with a neural capacity of 12 billion parameters. His alter ego, Midjourney, is a particularly stunning image generator, developed by an independent research laboratory. It has been available in open beta since July 2022.
Generative AI, like all AI, is based on technologies that have existed for quite a while, such as deep learning; artificial neural networks. In other words, artificial intelligence has neurons that are connected to each other in the same way that a human has neurons that are connected to each other using synapses.
The GPT1 to GPT-2: a state of infancy
Evolution from GPT-1 to GPT-2 (Screenshot from medium.com)
The acronym GPT translates into English as Generative Pre-Trained Transformer. Indeed, ” Generative » by the fact that these AIs are developed to generate text in natural human language; “Pre-Trained» reflects the idea that these AIs have been trained on gigantic quantities of data so that they can learn to find correlations between words, to recognize images in order to generate other content; and “Transform”, this technology translates the machine learning architecture on which AI is based.
Chat GPT is part of a large lineage of the family of generative artificial intelligence models developed by the American company Open AI, led by Sam Altman. Indeed, the first generative AI of this large family is GPT-1. It dates back to 2018. It was trained on a gigantic corpus of text from the internet with the aim of learning language structures and generating coherent textual content. However, GPT-1 was extremely limited in its ability to understand and generate text.
In February 2019, GPT-2 was released. This has seen significant progress compared to its predecessor. GPT-2 was trained on an even larger dataset. She demonstrated impressive text generation skills. However, it was still in its infancy as the results were not yet there. We therefore had to wait for the arrival of GPT-3 to begin to experience the thrills caused by the generative AI developed by Open IA.
GPT-3 to GPT-4: the pinnacle of Open AI’s generative AI
(Screenshot)
If the birth dates of the different Open IA babies are relatively close, the differences between them are remarkable. This is the case between GPT-1 and GPT-2. And it is even more phenomenal between GPT-3 and GPT-4. And in all of this, among the siblings, there is the giant, the one who is not the eldest who allowed us to put the spotlight on the large family of generative AI, Chat GPT.
On June 11, 2020, Open IA announced that GPT-3 is available to users. The latter is seen as a major breakthrough in the field of artificial intelligence. This is the start of a revolution for the world of generative AI. Speaking about GPT-3, the head of Open IA said: « GPT-3 was the first experiment where we really felt the intelligence of the system. He could do what a human can do. I think it got people who didn’t believe in artificial general intelligence at all to take it seriously. Something was happening that none of us predicted. » In order to avoid any confusion, taking into account the words of the boss of Open IA, it must be emphasized that there is a very big difference between the generative artificial intelligence that we tried to define a little above and the acronym breaking latest news which defines itself as “Artificial General Intelligence”.
Bluntly, we can state in the same vein that GPT-3 is really the model that put Open IA ahead in the race for the development of generative artificial intelligence. In an article published in the famous British daily, The Guardian in 2020, GPT-3 wrote an essay, entitled: “Are you still afraid, human? » In this essay, GPT-3 tries to convince humans that it is there for the well-being of all of them. However, while GPT-3 promises not to be there to exterminate humans; she continued her essay by stating: “Humans must continue to do what they know how to do, to hate and fight each other. I’ll sit in the background and let them do their thing. » This statement could suggest that AI will not change anything in human relations. But to understand the evolution of current things; it’s quite the opposite. To the extent that, particularly new technologies, AI has completely disrupted human relations from an economic, political, legal, social and cultural point of view.
While GPT-3 put Open IA at the front of the race to develop generative AI, the race did not stop there. On November 30, 2022, Open IA reached its peak with the release of its Chat GPT conversational agent prototype. The latter is more powerful than the first generative AI launched by Open IA. This is explained by the fact that Chat GPT was trained on much more data compared to its predecessors. In addition, Chat GPT includes 175 billion parameters; unlike GPT-3 which includes only 1.5 billion parameters. All this explains the power of Chat GPT. We could say that it has been better nurtured than its predecessors if we take into account the fact that data remains and remains the holy grail of AI in general and particularly of generative AI.
Since its launch, Chat GPT has continued to attract attention. Some speak of him in glowing, optimistic terms. This category of people sees the conversational agent as a revolution. Which is difficult to question given the feats that Chat GPT has achieved. However, others are not so happy with the chatbot; because many people see it as an imminent danger, a threat to the future of humanity. According to this second category of people, the challenges that generative AI represents on the socio-economic, political, legal and cultural levels are too enormous to glory in this tool. Opinions differ regarding the continued development of generative AI. But this did not stop Open IA in the development of its generative AI. Thus, on March 14, 2023, Open IA gave birth to the latest GPT-4 baby.
GPT-4: barely visible differences to GPT-3
For some the difference between GPT-4 and GPT-3 is barely visible apart from the fact that you can perform searches by integrating images into the input with GPT-4. According to Open AI, the difference between the two models reveals itself when it comes to accomplishing a task with a certain complexity. At a certain threshold of complexity, GPT-4 proves to be more reliable and more creative. In addition, it is able to take into account instructions with a certain level of nuance. However, we do not know the quantity of parameters and the number of data ingested by it. But one thing is certain, it contains more parameters and data than the previous ones.
Since the launch of Chat GPT, many companies in the technology world are developing their own generative artificial intelligence. This is the case with the giant Google which launched Bard on March 21, 2023; and Microsoft, which launched Bing AI, its conversational robot, in February 2023. Furthermore, Baidu, the tech giant from the Middle Kingdom, tried to follow suit in the world of generative AI, but its launch was judged above did not live up to expectations. Many other tech companies, for the sake of competition, have tried to compete with Chat GPT. But until then, Chat GPT remains relentless in the generative AI scene.
If the prowess of generative AI is seen as a revolution in the field of artificial intelligence, the stakes remain high. Thus, it is important to question its socio-economic, political, legal, ethical and cultural issues; and in particular on the increase in false information commonly called fake news in our current societies. The truth is that these multiple technological advances do not remain without consequences on the future of societies in general and on the future of so-called democratic societies in particular.