Yann LeCun has been a professor of computer science at New York University since 2003, and he came to prominence because of his excellent work on convolutional neural networks (CNNs). In addition to being the Chief AI Scientist at Meta, he is the most level-headed of the world’s AI celebrities. He isn’t terribly hopeful or terribly afraid of AI’s potential, and he expressed his rational views in two recent discussions.

At the 2024 K-Science and Technology Global Forum (reported here), he explained how limited LLMs are compared to human intelligence. He said that LLMs “cannot deal with the complexity of the real world”, and that they “lack the ability to reason, plan, and understand the physical world in the same way humans do.” He also stated that no AI system has been shown to be inherently dangerous, and that open-source models should be embraced with a minimum of regulation.

During Nikhil Kamath’s podcast (reported here), Prof. LeCun further presented his thoughts on open-source models in general and Llama in particular. He stated that “The proprietary engines will not be nearly as important as they are today… a fine-tuned open-source engine like Llama always works better than a non-fine-tined generic top-performing model.”

Reporters have paid a great deal of attention to prophets of fear like Sir Geoffrey Hinton, who champion government regulations and proprietary models like OpenAI’s GPT. It’s a shame that sensible researchers like Prof. LeCun get overlooked.