While large language models like ChatGPT may seem really smart, they lack key capabilities along with many important dimensions, Yann LeCun — executive chairman of Advanced Machine Intelligence Labs and professor of computer science at New York University — said at Brown’s 2026 Lemley Family Leadership Lecture.
“In fact, they are completely helpless when it comes to the physical world,” LeCun said.
On a slide shown during his presentation, LeCun encouraged fellow artificial intelligence scientists to “abandon” generative models and other well-established methods. In all-caps and red text, the slide read: “IF YOU ARE INTERESTED IN HUMAN-LEVEL AI, DON’T WORK ON LLMs.”
LeCun feels much more confident in a different type of artificial intelligence.
“There is no question … that machines will eventually surpass humans in all domains where humans are intelligent,” he said, though he noted that this will take “a while” and is “almost certainly much harder than we think.”
LeCun’s company, AMI Labs, is an AI startup that is developing the concept of “world models,” which are AI models able to understand the physical world because they are trained with sensory data, as opposed to language-based models, which are trained on human-produced texts. The company has raised over $1 billion in seed funding.
“We have various systems that can write code, they can pass the bar exam, they can win international math Olympiads,” LeCun said. “But where is my domestic robot? Where is my robot that can clean the house, learn to drive in 20 hours of practice?”
LeCun defines intelligence as “the ability to accomplish new tasks you’ve never been exposed to and solve new problems without any prior training” — a skill he says LLMs do not possess. He argued that because of this, LLMs will not be able to reach human intelligence, saying that “human-level AI will require real world data, sensory inputs, as opposed to just language or text.”
AMI Labs’s world models are still in the early stages of development. According to LeCun, these models will be trained with sensory data so that they can understand the physical world and “predict the consequences of their actions in advance.”
Nearly all of the current AI systems are incapable of predicting their actions’ consequences, LeCun said. He described current LLMs as “intrinsically unsafe.”
But with world models, “guardrail objectives” can be implemented, which means that “by construction, they will not knowingly produce actions that will produce dangerous results,” he added.
Although world models bring a host of technical processes and challenges to work out, LeCun said that the work he has been doing over the past 15 years is “enabling the next AI revolution.”
One of the key challenges is working out hierarchical planning, the ability to prioritize actions in a sequence, and a hallmark skill of human intelligence.
“Here is the big secret of AI: Nobody knows how to do hierarchical planning,” he said.
He said his goal for AMI Labs is “to become the main provider of intelligence systems,” a goal he joked is “very modest, not ambitious at all.” He added that these new AI systems have potential applications and value “in all major sectors,” including the economy, academic fields, the physical and medical sciences and more.
In the Q&A session, LeCun emphasized the impact of academic research on industry and the importance of collaboration. He said he is “a firm believer” that good ideas “come from the interactions between people working on different assumptions with different motivations in different environments.”
When asked about his view on education in the current world of increasing technological and AI advancement, LeCun said that the sentiment that education is useless with an increased reliance on AI is “not true,” noting a trend in increased demand for more advanced degrees, specifically in STEM fields.
He also advised students to “study things that have a long shelf life and are very fundamental,” noting that “technological progress is accelerating” and “you’re almost certainly going to have to change jobs during your career.”
Attendee Eduardo Michelsen ’29 said it was “an honor to have someone like (LeCun) on campus,” calling the talk “insightful.”
“He made it sound like at least studying something like (computer science) was probably something smart to do, even though it seems like it’s not,” Michelsen said.
William Yu ’26, who also attended the event, said he thought the lecture was “probably the coolest event, most interesting speaker that (he has) listened to in (his) time at Brown.”
“He believes in the direction he’s going in with world models, and he’s not afraid to say that he thinks LLMs are a dead end,” he said. “I think he has earned the right to be clear in what he thinks with the contributions he’s made to the field.”
Yu appreciated the talk’s connection to how AI will affect students in their future everyday lives.
“I think it’s very important that the University brings these kinds of voices to campus and gets more students informed about these changes that are happening,” he said.
Rachel Wicker is a senior staff writer covering affinity and identity. She is from Athens, Georgia and plans on concentrating in English on the nonfiction track and International and Public Affairs. Outside of writing, she enjoys reading books of any genre and doing yoga.
