Yann LeCun, Meta’s Chief AI Scientist, Turing Award winner, and a professor of Computer Science at New York University, has reignited an important debate in late 2025 about how students should prepare for the future of artificial intelligence. Contrary to some headlines, LeCun is not advising students to abandon computer science altogether. Instead, he is urging them to stop chasing short-lived tech trends and refocus on foundational disciplines that will remain valuable as AI continues to evolve.
Why LeCun Thinks Traditional CS Is Not Enough
LeCun argues that many modern computer science programs are becoming overly focused on tools of the moment specific programming languages, frameworks, or even current LLM prompt engineering techniques. While these skills may offer quick job opportunities, they often lack long-term durability.
According to LeCun, routine coding is increasingly being automated by AI itself. As systems grow more capable, engineers whose primary value lies in writing standard application code may find their skills rapidly commoditized.
What Students Should Study Instead
LeCun strongly recommends subjects with what he calls a “long shelf life” knowledge that remains useful regardless of shifts in technology:
Advanced Mathematics
He believes most CS degrees fall short in math requirements. LeCun advocates mastering Calculus I, II, and III, Linear Algebra, Probability, and Statistics, arguing that these areas form the backbone of modern AI and machine learning.
Physics
Physics plays a central role in LeCun’s thinking because it teaches students how to model the real world, understand complex systems, and identify which details matter. These skills are critical for building AI systems capable of true reasoning and “world modeling.”
Engineering Foundations
Courses in Electrical Engineering, signal processing, and control theory are also key. These disciplines provide practical frameworks for understanding how information flows and how systems interact with real-world constraints.
Modeling Reality
LeCun emphasizes learning mathematics that connects directly to reality rather than purely abstract symbol manipulation a requirement for building the next generation of intelligent machines.
The Bigger Problem With Today’s AI
LeCun believes today’s Large Language Models are hitting a ceiling because they are essentially next-token predictors. While powerful, they lack common sense and physical understanding, limiting their ability to reason about the real world. Overcoming this requires deep knowledge of math, physics, and engineering not just better prompts.
A Career Strategy for the Long Term
One of LeCun’s most quoted lines captures his philosophy perfectly:
“If you have to choose between iOS programming and Quantum Mechanics, take Quantum Mechanics.”
Frameworks can be learned on the job. Foundational thinking cannot.
LeCun’s Own Academic Path
LeCun often highlights that he was not trained as a computer science engineer. He studied Electrical Engineering at ESIEE Paris before earning his Ph.D. He credits this background for enabling his breakthrough work on Convolutional Neural Networks (CNNs) the foundation of modern computer vision.
