Yann LeCun is famously pessimistic on whether LLMs can get us to AGI, but he also seems to be pessimistic about the recent advances in humanoid robotics.
The Meta Chief AI Scientist, a Turing Award winner known for his pioneering work on deep learning and convolutional neural networks, has raised pointed concerns about the wave of humanoid robotics startups that have emerged in recent years. His comments cut to the heart of a fundamental challenge facing the industry: the gap between impressive hardware demonstrations and genuine artificial intelligence that can handle diverse, real-world tasks.
“There is a large number of robotics companies that have been created over the last few years building humanoid robots,” LeCun noted. “The big secret in the industry is that none of those companies has any idea how to make those robots smart enough to be useful, or I should say smart enough to be generally useful.”
His critique acknowledges that current approaches have their place, but within narrow boundaries. “We can train those robots for particular tasks, maybe in manufacturing and things like this, right?” he said. “But your domestic robot—there is a bunch of direct tools that need to arrive in AI before that’s possible.”
The problem, according to LeCun, isn’t just incremental—it’s foundational. “So the future of a lot of those companies essentially depends on whether we’re going to make progress, significant progress towards those kind of world model planning type architectures,” he explained, pointing to the need for breakthroughs in how AI systems understand and reason about the physical world.
LeCun’s assessment arrives at a critical moment for the humanoid robotics industry. Companies like Figure AI, 1X Technologies, and Tesla with its Optimus robot have collectively raised billions in funding, often showcasing sleek prototypes performing tasks like folding laundry, walking through warehouses, or manipulating objects. Yet LeCun’s comments suggest these demonstrations may obscure a harder truth: today’s AI systems lack the fundamental capabilities—such as robust world models and genuine planning abilities—needed for robots to operate reliably across the unpredictable scenarios of everyday life.
The implications are sobering for an industry banking on near-term commercial viability. If LeCun is correct, many of these well-funded ventures may be building sophisticated hardware waiting for AI breakthroughs that remain years or decades away. Unlike manufacturing environments where tasks are repetitive and controlled, domestic and general-purpose applications demand robots that can adapt to novel situations, understand context, and plan multi-step actions in dynamic environments—capabilities that current AI architectures, despite their impressive advances in language and narrow robotics tasks, have yet to demonstrate convincingly. The robotics boom may be running ahead of the intelligence needed to power it.