OpenAI and Meta have announced their new artificial intelligence models, GPT-5 and Llama 3, promising significantly improved “reasoning” capabilities. These models, still in development, aim to perform complex tasks such as planning and booking trips, solving “difficult” problems, or even speaking.
AI reasoning, according to experts
Joelle Pineau, vice president of AI research at Meta, emphasizes the importance of these advances: “We are working hard to get these models to not only speak but also reason, plan, and even remember."”
For his part, Brad Lightcap, director of operations at OpenAI, talks about GPT-5's improvements in terms of solving "difficult problems," a considerable evolution compared to previous versions.
Implications of AI reasoning
At an event in London, Meta artificial intelligence researcher Yann LeCun criticized the limitations of current AIs, which produce "one word after another without thinking" and "always make stupid mistakes.""
The reasoning ability of new models could radically transform their applications, allowing, for example, the complete planning of a trip with an understanding of the implications of each step.
Towards artificial general intelligence (AGI)
These advances raise the question of the arrival of artificial general intelligence (AGI), an AI capable of cognitive performances comparable to or superior to those of humans in a wide variety of fields.
OpenAI envisions “highly autonomous” AI that outperforms humans at the most economically profitable tasks.