The new model of the American company was eagerly awaited when the race for AS assistants always more efficient is in full swing between the technologies giants, with a target of an AI called “general” or “superintelligent”, with cognitive capacities higher than those of humans.
“GPT-5 is like talking to a doctoral level expert on any subject,” assured Sam Altman, boss and co-founder of Openai, during a breff for the press.
He compared GPT-3, at work in the first version of Chatgpt, to a high school student who “sometimes gives the right answer, and sometimes responds anything”, and GPT-4 to a university student.
Companies at the top of the generative AI race compete with new models that are ever more sophisticated in terms of capacity to “reason” and do tasks independently.
The pace has accelerated because the different groups seek to make their assistant IA – Chatgpt, Gemini (Google), Meta Ai, Claude (Anthropic) etc. – Indispensable on a daily basis for the greatest possible number of users and developers.
“Super power”
Google and Meta (Facebook, Instagram) capitalize on their immense user bases. Anthropic has made a name for himself especially with professionals. Grok, from Xai, an Elon Musk company, is integrated directly into the social network X.
The Chinese start-up Deepseek created the event at the start of the year by leaving R1, an open and advanced model despite technological and budgetary constraints.
But Chatgpt remains the best known name in the general public. It has nearly 700 million weekly active users and Openai intends to maintain its lead.
The assistant will become customizable – the user can choose a concise, benevolent or sarcastic tone – and will be able to connect to Gmail.
To see too | Artificial intelligence as a psychologist?
“GPT-5 can really do impressive things for you. It can instantly create software, on demand (…) It is a great incredible power, ”praised Sam Altman.
One of these engineers, Yann Dubois, asked the Assistant IA in everyday language to create an online application to learn French with games. GPT-5 immediately produced hundreds of lines of code and the corresponding, elementary website was created in a few minutes.
According to Openai, GPT-5 is also more reliable: he “hallucinates” much less than the previous models, and he “admits” when he does not know instead of inventing an apparently convincing response.
“We led him to be honest (…) so that he does not lie to users,” said Alex Beutel, a company security manager.
“Do not harm”
In terms of security, uses for harmful purposes are also becoming more difficult.
“Before, the approach was binary. If the request seemed safe, the model was running, and if not he refused, “said the researcher.
With GPT-5, in case of doubt about potential criminal motivations, “the model will be limited to general information, which cannot harm”.
Openai was founded as a non -profit organization in 2015, with the mission of creating a general AI which benefits all humanity.
GPT-5 “is clearly an intelligent model, but it lacks something important to reach general AI (…) It is not a model that continues continuously from things that it discovers,” said Sam Altman.
“Obviously, we will have to invest in the computation power with dizzying costs to get there, but we intend to continue on this path,” he added.
The Silicon Valley Star Star Star, who decided this year to remain for non-profit after many debates, raised 40 billion dollars in March, which valued it to 300 billion.
Microsoft, the main Openai, Google, Meta and Amazon investor, have planned to spend each tens of billions of dollars in IA infrastructure (sophisticated data centers), cutting-edge chips and ultra-qualified human resources this year.