ChatGPT revolutionizes the field of AI. © Getty Images
Opened to the public on November 30, the conversational agent ChatGPT impresses as much as it worries. According to industry experts, the language model used by artificial intelligence, GPT-3, constitutes a real technological revolution and opens the way to an infinite field of possibilities. Today we explain the basics of this little digital being which could well be the innovation of the year.
ChatGPT, what is it exactly?
The easiest way to answer the question is to ask the system itself.
“I am a computer language model designed by OpenAI. My goal is to be able to answer questions and provide help to users using my machine learning and natural language skills. I try to provide accurate and useful answers to users. I am a natural language processor and have no personal opinions or preferences“, then replies ChatGPT.
The principle is simple, the user asks a question and the conversational intelligence tries to answer it with its acquired knowledge. A more or less detailed answer is generated, in a language almost identical to that of a human. This is the strength of this new service: being able to formulate an answer in the same way as a loved one would.
What can AI actually do?
ChatGPT is content to answer the questions and problems of the user. And it’s already huge. All the little thinking tasks of daily life can be subjected to AI. A complex mathematical operation, a simple logic problem, a history lesson… Students and other school children will not fail to use the services of the chat while doing their homework.
But the most impressive functions lie in the generation of bespoke text. ChatGPT is able to create a fictional story from scratch, develop a simple script for a video, translate a text, or even write articles on given topics.
On a day-to-day basis, conversational intelligence can create a specific diet, dedicated sports training, or diagnose certain illnesses based on symptoms. Possibilities which cannot replace a consultation with a professional, reminds the IA on several occasions.
On the side of new technologies, the arrival of this bot could well be a game changer. ChatGPT is particularly capable of developing, in complete autonomy, pieces of code in a slew of programming languages. AI can also analyze scripts to detect bugs or vulnerabilities.
ChatGPT generates a program in seconds. © Capture ChatGPT
ChatGPT is based on the GPT-3 language model. The AI trained in the basics of several languages using the deep learning, or deep learning in French. A textual corpus of considerable mass has been subjected to algorithms to develop a common sense of language.
AI trainers have also “provided conversations in which they played both roles, a user and an AI assistant“, explains OpenAI. And to detail: “We randomly selected a post written by the model, sampled several alternative responses, and asked the AI trainers to rank them. Using these reward models, we can refine the model using proximal policy optimization.“
As for pure knowledge, ChatGPT relies on a database of texts present on the web in 2021. A certain limit that sometimes prevents the robot from answering certain questions correctly. Part of the Wikipedia database is thus indexed in the memory of this digital titan.
Who is behind ChatGPT?
Behind ChatGPT hides, again and again, the whimsical billionaire Elon Musk. Founded as a non-profit association in 2015 by Elon Musk and Sam Altman, OpenAI specializes in the development of artificial intelligence of all kinds. The organization is also behind the stunning image-generating AI Dall-E.
The stated goal of OpenAI is to “ensuring that artificial general intelligence (AGI) — by which we mean highly autonomous systems that outperform humans in most economically worthwhile work — benefits all of humanity.“
What are the dangers of this technology?
Although ChatGPT is, for the time being, more beneficial than harmful for humans, several major problems can arise. The main weakness of AI lies in the limitation of its knowledge. As learning ended in 2021, AI knowledge is skewed. Answers requiring knowledge after that year should be taken with a grain of salt. In some cases, ChatGPT won’t hesitate to lie, and give a completely false answer.
Thus, when the agent is asked about the name of the current boss of Twitter, the latter answers without flinching “Jack Dorsey“. However, the latter has not been in control of the social network since November 29, 2021. The indexing of ChatGPT seems not to have taken the event into account. Only a connection of the bot to the Internet could, in appearance, resolve this gap.
False information. © Capture ChatGPT
Another problem: by allowing the generation of code, AI is able to develop complex programs, which once took several hours of thought for a programmer. ChatGPT therefore seems able to produce malware in a few seconds. Similarly, AI can detect a vulnerability by simply submitting source code. A tremendous time saver for potential hackers.
Properly used, however, this technology could accelerate the securing of complex computer systems.