Yann LeCun, Meta’s former chief AI scientist and a Turing Award laureate, has issued a stark warning to the technology industry and investors banking on the imminent arrival of human-level artificial intelligence: current methods are insufficient to get us there.
Speaking on the Big Technology Podcast with host Alex Kantrowitz, LeCun dismantled the prevailing narrative that simply feeding more data and computing power into Large Language Models (LLMs)—the technology underpinning ChatGPT and Meta AI—will result in Artificial General Intelligence (AGI).
“We are not going to get to human-level AI by just scaling up LLMs. This is just not going to happen,” LeCun said. When pressed on whether this was merely a difference of opinion, he doubled down: “There is absolutely no way.”
The Limits of Current Tech
LeCun’s critique centers on the fundamental architecture of today’s generative AI. While acknowledging that these systems have “gigantic memory and retrieval ability,” he argues they lack the cognitive architecture required for true intelligence.
“It would feel like you have a PhD sitting next to you,” LeCun said, describing the utility of current assistants. “But it’s not a PhD you have next to you… [it’s] not a system that can invent solutions to new problems.”
He dismissed the optimism of some of his “more adventurous colleagues” regarding timelines, stating that AGI “is not going to happen within the next two years.” He argues that LLMs inherently lack persistent memory, reasoning, planning, and an understanding of the physical world—four pillars he believes are essential for human-level intelligence.
The Multi-Billion Dollar Gamble
The interview touched on the disconnect between LeCun’s skepticism and the massive capital expenditure currently underway in Silicon Valley. Tech giants, including Meta, Microsoft, and Google, have been pouring tens of billions of dollars into data centers and GPUs.
LeCun clarified that this investment was pragmatic rather than speculative. He noted that Meta’s goal was to have “one billion users of Meta AI” across its platforms. The infrastructure is required to serve these users today, regardless of whether a philosophical breakthrough occurs tomorrow.
“The investment that is being done now is not done for tomorrow… it’s done for the next few years,” LeCun explained, adding that “you can’t afford not to do it” given the competitive landscape.
Risks of an ‘AI Winter’
Kantrowitz raised the issue of reliability, noting that for enterprise clients, an AI that is 95% accurate is often useless due to the remaining 5% of hallucinations. LeCun agreed, drawing parallels to the hype cycles of the past, specifically the excitement surrounding IBM Watson and the “expert systems” of the 1980s. Both technologies attracted massive interest and funding before failing to meet lofty expectations, leading to periods of reduced funding and interest known as “AI winters.”
LeCun cautioned investors against believing that a small startup has secretly solved AGI.
“If you think that there is some startup somewhere with five people who has discovered the secret of AGI and you should invest five billion in them, you are making a huge mistake,” he said. “It’s not going to come from a single entity, it’s going to come from the entire research community.”
The Path Forward
According to LeCun, the path to true intelligence requires a shift away from text-based training toward systems that can learn from video and real-world interactions to build “mental models” of the physical environment. He estimates that significant progress toward systems that can reason and plan may occur within “three to five years,” but stressed that it will be a gradual evolution, not a singular event.
“This is just not going to happen,” LeCun reiterated regarding the pure scaling hypothesis. “There is absolutely no way in hell.”