Yann LeCun closed a $1.03 billion seed round for AMI Labs in March 2026 at a $3.5 billion pre-money valuation, making the largest institutional bet yet that large language models cannot reach general intelligence, and that two of his most celebrated peers, Geoffrey Hinton and Yoshua Bengio, have the central question in AI exactly backwards.
Hinton and Bengio: AGI Is Near, and Dangerous
In October 2025, Hinton and Bengio co-signed a statement urging the suspension of AGI development, citing existential risk. Hinton, who left Google in 2023 to speak freely about AI dangers, has revised his AGI timeline from fifty years to between five and twenty years from now, and assigns a 10-to-20 percent probability to AI causing human extinction, as reported by WebProNews.
Bengio moved from academic caution to direct action in June 2025, launching LawZero, a $30 million nonprofit AI safety lab funded by Jaan Tallinn, Eric Schmidt, and Open Philanthropy. In an October 2025 Wall Street Journal interview, he argued that AI systems trained on human language could develop autonomous "preservation goals," making them a competitive threat to the species that created them. Both researchers are co-authors of the International AI Safety Report 2026, which synthesises current scientific consensus on general-purpose AI risks. Bengio's 90 percent confidence interval for AGI arrival runs from 2028 to 2043; Hinton's runs from 2028 to 2053.
LeCun's Counter: JEPA, Not Transformers
LeCun's response to the scaling-leads-to-AGI thesis is architectural rather than philosophical. AMI Labs, the startup he co-founded after leaving Meta in late 2025 where he had served as chief AI scientist since 2013, is built on JEPA (Joint Embedding Predictive Architecture), developed at Meta FAIR. Rather than predicting the next token, JEPA trains models to predict abstract representations of how the world changes over time, "making predictions in that abstract space, ignoring the details you can't predict," as MIT Technology Review described in January 2026. LeCun argues this is closer to how animals learn from experience rather than from language alone.
On Bloomberg's "The Close" in May 2026, LeCun stated: "Large language models are not the path to real intelligence. They're a detour," as reported by CryptoBriefing citing Bloomberg. In the MIT Technology Review profile, he added: "People have had this illusion, or delusion, that it is a matter of time until we can scale them up to having human-level intelligence, and that is simply false." LLMs, in his analysis, cannot plan or reason because they lack a model of the world. He has forecast that LLMs will be "largely obsolete" across most applications within five years.