Meta scientist doubts artificial intelligence will rise to human or superhuman levels

By Kate Irwin

To some extent, an AI model is only as “smart” as the data it’s trained on — and Yann LeCun, Meta’s Chief AI Scientist, argues that large language models (LLMs) like ChatGPT, Google’s Gemini, or Meta’s Llama won’t ever reach human levels of intelligence.

Meta Chief AI Scientist Yann LeCun

The concept of an AI model achieving human or superhuman intelligence is often called artificial general intelligence. AGI doesn’t exist yet and remains a hot topic among AI researchers and executives. While some, like OpenAI CEO Sam Altman, are convinced AGI is coming, others, like LeCun, are less sure when or how this might happen.

LLMs have a “very limited understanding of logic,” cannot comprehend the physical world, and don’t have “persistent memory,” LeCun tells the Financial Times. While OpenAI recently gave ChatGPT a kind of “working memory,” LeCun doesn’t think current AI models are much smarter “than a house cat.”

“It certainly appears to most people as reasoning—but mostly it’s exploiting accumulated knowledge from lots of training data,” LeCun told FT of existing LLMs. They’re also “intrinsically unsafe” because they rely so heavily on the training data, which could contain inaccuracies or be out-of-date (AI models are still prone to hallucinating fake information).

Last week, LeCun gave a proverbial eye-roll to OpenAI researcher Jan Leike, who resigned from OpenAI and expressed concerns over a supposed urgent need to “figure out how to steer and control AI systems much smarter than us.”

“It seems to me that before ‘urgently figuring out how to control AI systems much smarter than us’ we need to have the beginning of a hint of a design for a system smarter than a house cat,” LeCun wrote on X, adding: “Such a sense of urgency reveals an extremely distorted view of reality.”

Leike was part of OpenAI’s Superalignment team, which was intended to prevent AI from “lead[ing] to the disempowerment of humanity or even human extinction.” With the departure of Leike and fellow team lead Ilya Sutskever, that team has disbanded, Wired reports.

At Meta, LeCun is focused on developing an entirely new type of AI system that isn’t like current LLMs. He expects this could take up to a decade to materialize fully, according to FT‘s report. LeCun’s team consists of about 500 Meta employees who are working with the common goal of trying to develop an AI with common sense and a true sense of how the world works—instead of chatbots that spit out garbled images of human faces, landscapes that defy physics, and bizarre illustrations of people with ten fingers on each hand.

“We are at the point where we think we are on the cusp of maybe the next generation AI systems,” LeCun said.
_________________

Credit: Financial Times