The head of Meta (Facebook, Instagram) believes that generative artificial intelligence – the technology behind ChatGPT – has already reached a dead end, and promises a new artificial intelligence closer to human, reports AFP.

ChatGPT is a tool that can make your life and work easier if you know how to use it properlyPhoto: Dmytro Melnikov/Alami/Alami/Profimedia

“Artificial intelligence and machine learning are really ugly today. Humans have common sense, machines don’t,” Jan LeCun, Meta’s chief artificial intelligence scientist, told reporters on Tuesday.

He made the announcement at an event in Paris, where the California group had just presented a project for a new artificial intelligence architecture called the Joint Image Embedding Predictive Architecture (I-JEPA).

This computer vision technology “learns (…) by creating an internal model of the external world by comparing abstract images (rather than comparing the pixels themselves),” Meta explained.

Currently, generative artificial intelligence programs such as ChatGPT or Bard (Google) rely on linguistic models trained by huge databases to be able to predict which word follows another, all the way to producing any texts (essays, speeches, poems, etc. ).

Programs such as DALL-E or Midjourney create images using the same principle.

“This is already a revolution,” admitted Yann LeCun. “If you train (a model) with 1,000 or 2,000 billion chips, it seems to be able to understand. But he makes silly mistakes, be they factual or logical.”

“Generative methods tend to focus too much on details instead of embracing global, predictable concepts,” Meta said in a press release.

I-JEPA, which is based on the vision of a French scientist, is designed to allow a machine to think abstractly, more like a human.

It will be open source, meaning it will be open to researchers who want to test it.

The goal is to develop artificial intelligence that “better reflects the way people understand the world,” Mark Zuckerberg, Meta’s boss, said on his Facebook profile.

“Generative models are a thing of the past, we will abandon them in favor of predictable architectures,” such as the one presented by the American group, assured Ian LeCun.

He also promised a “new rebirth of humanity” thanks to AI.