AI Leaders LeCun and Fayyad Cut Through ChatGPT Hype, Chaos and Confusion in a Fireside Chat

LeCun boldly claims, “Machine Learning Sucks.”

Legendary AI pioneer Yann LeCun laid out a bold vision for the future of artificial intelligence in a Distinguished Lecture Seminar and fireside chat hosted by the Institute for Experiential Artificial Intelligence (EAI) at Northeastern University.

The sold-out event, which featured a lively exchange between LeCun and the Institute’s Executive Director, Usama Fayyad, was well-timed to counter confusing narratives surrounding generative AI tools like ChatGPT. It also served to demystify the technology and show how it can be useful to businesses, educators, and workers across all industries. The presentation by LeCun, who is a Turing Award winner and currently serves as chief AI scientist at Meta AI (FAIR), also highlighted the need for an approach to AI that relies on human intervention.

“Technologies like large language models (LLMs) are amazing in terms of the performance they produce…but they make a lot of factual errors, logical errors, have inconsistencies, have limited reasoning abilities, and they’re pretty gullible,” LeCun told a packed house at ISEC Auditorium. “We need to figure out how to get machines that can not only learn but also reason and plan. Current AI systems cannot do this.”

LeCun said five years from now, no one will use LLMs for research purposes. Fayyad said the message — which goes against the media narrative — is a sorely-needed addition to today’s discourse. Beyond the media hype, Fayyad believes the economic impact of LLMs will far outlast their research usefulness and says LLMs will accelerate many knowledge worker tasks.

“The buzz is at crazy levels and hype is everywhere,” Fayyad said. “That’s why it’s important for EAI to bring in these views from people pushing the state of the art of the technology, who can say, ‘Don’t be fooled.’ Yann helped [advance these systems], but now he’s at a point of disillusionment, to where he’s saying things like “machine learning sucks,” because the systems don’t even understand what they’re doing.”

LeCun concluded his presentation by describing a self-supervised AI system that uses a hierarchical world model and hierarchical planning to achieve more human-like reasoning. Fayyad called the approach a much needed refresh to the way research in deep learning has been done – with almost zero domain or reasoning knowledge.

The fireside chat portion of the event gave guests a chance to hear LeCun and Fayyad’s thoughts on a wide range of issues as the two dug their way into over 150 audience questions about where AI is going and how it will impact society.

"What would you have told Congress in the hearing featuring OpenAI CEO, Sam Altman?” Fayyad asked LeCun.  

“I’d tell Congress not to ban open source LLMs,” LeCun responded. “They’re not a threat to humanity. Yes there are bad actors, but make it open, it’s the only way to make it safe."

Fayyad said the talk came at the right time and exemplifies the Institute and University’s goal of bringing AI leaders together for critical conversations.

“The concept Northeastern University President Joseph Aoun talks about in his book — the need to become ‘robot proof’ — is more relevant than ever with what we've seen from generative AI in the last year or so,” Fayyad said. “The event had the right timing and the right philosophy, which is consistent with the university’s strategic plan to go into the real world and find real problems to work on. That’s why the Institute for Experiential AI has been getting so much interest from industry. It’s also why we saw so much interest from students at the event — because they all recognize the value of experiential learning.”

Watch the full talk here. Stay tuned for part two, which will feature answers from LeCun and Fayyad to many more of the hundreds of audience-submitted questions.