The world was getting used to the functionalities of phone tree to help customers for example, but nobody was expecting the surge of AI in chatbots as it is happening. The private company OpenAI created the effect of a tsunami with different versions of ChatGPT. The observers were reluctant to shout at a genius solution, till April 2023. GPT stands for Generative Pre-trained Transformer. The first mention of this type of Artificial intelligence appeared in a paper in 2018. Since the pattern is based on data for the internet, it did not catch a lot of attention nor big concerns in the public opinion. Surprisingly, the model quickly evolved in two other versions, upgrading the first model. The pace is incredibly fast : GPT-2 appeared in 2019 and GPT-3 in 2020.
What the different versions of GPT are revealing?
It is important to follow the historic development of the communication of the performances of the GPT to understand the revolution we are going through. In November 2022, the Californian based company developing GPT made public a version of GPT called ChatGPT, which was based on GPT 3-5. As early as April 2023, the world heard about ChatGPT 4 and its extraordinary capacities. There were public announcements explaining about the improvements of the technology, but also a lot of outcries. It is expected when one announce improvements in technology to receive criticisms. The fight of traditions and technical innovation is nothing new. From the dawn of the Industrial revolution, people fought to keep their jobs and to stop the evolution of technology with concerns for the environment. Never before the scale of the threat to human beings has been of this magnitude.
There are two ways to react to chatGPT4:
- to analyse clearly the functionalities and to understand the performance
- to shout and predict the end of a certain world.
Actually, it is true to admit that, since the use of the ChatGPT is allowed for free, it seems unstoppable. Sam Altman, the CEO of OpenAI, the company behind ChatGPT, became a worldwide celebrity in no time. It is delicate to qualify with words every progress of the GPT system, since every step seems to improve dramatically the capacities of the machine versus the human. Truth to be told, GPT4 is moving forward at a great speed towards the building of a surprisingly clever machine.
Why is GPT-4 the better version?
GPT 4 is a multimodal model. It means it treats the inputs differently. It accepts not only texts, but also images. This enables the machine to describe for example what is funny in pictures. This is not always a necessity, and even if it sounds a minor performance, it means the machine is getting closer to human capacities to distinguish hidden meanings in images. In other words, the analysis is outperforming the previous versions of GPT, at an alarming speed for some, not fast enough for others. What is certain, is the creation of new perspectives for people working using their intellectual capacities. Manual workers are not concerned nor threaten by the use of a ChatGPT-4 in their jobs. So far, any profession developing soft skills is not concerned either. ChatGPT-4 is like an engine generating some kind of answers to some patterns of queries, but it cannot interact with humans from an emotional angle.
The limits of LLM’s and GPT 4
It must not be forgotten that as today in 2023, the model training the machine is functioning thankfully to a massive quantity of data sourced from the internet. This is not without creating difficulties. Markers are needed to operate a sort of cleaning to expel off the model wrong or distorted information. Consequently, the fascination and fears about the consequences of the development of GPT-4 as of today are not entirely grounded. But the examples of answers provided by the machine and widely communicated in the press are provoking growing concerns for what might happen in a not so remote future. For example, the answers of the machine are corresponding to correct answers for bar examination. The machine was able to use data and structure legal information in a logic way. One might wonder if the questions were too easy or from “old” examination papers and easy to rehash from the internet. Consequently, some enthusiastic people are claiming that consultants, doctors could soon be replaced by machines. This could sound like a Vanity Fair where the biggest prophecy will win.
Generative AI use cases
On the other hand, it is almost shocking to witness at which speed the machine can produce a novella with some indications or can summarize a text. Another big improvements of ChatGPT-4 is its capacity to align in a more efficient way what the user is expecting with what it is producing. It had been witnessed for the previous versions a capacity of the machine to “hallucinate”, that is to say, to imagine answers and to deny that it was imagined. Also, since it is possible to get anything on the internet, from scholars extract of thesis to the most abject affirmations about anything, the conception of ChatGPT-4 is already improving the fights against these problems but not solving the matter. So far, perfection is not attained. Nonetheless, some companies are already integrating ChatGPT-4 in their solutions to try to improve their performance. A famous app to learn languages, Duolingo, is using the AI for example to improve the answers provided to their users as feedbacks. This is still in the testing phase, but it is an interesting case to fully understand why all of a sudden several companies are investing money and resources to develop AI chatbots. The race to produce faster video games, answers, and information in finance, or description of the environment to the visually impaired has begun. Numerous companies all over the world are fighting to become the leader in a new market. In the past, the development of technology had been explained by the motivation to free the people from repetitive, demanding tasks to allow humans to be more creative. This is also an argument used by the companies developing AI chatbot.