Artificial intelligence may be dazzling the world with chatbots and image generators, but leading researchers say today’s systems are still far from truly intelligent especially when it comes to robotics. All information has been verified with trusted, up‑to‑date sources.
Despite billions of dollars pouring into humanoid robots capable of walking, lifting and navigating complex environments, teaching them to perform simple household tasks ironing, stacking a dishwasher, picking up a pen remains slow, expensive and unreliable. And according to Yann LeCun, one of the world’s most influential AI scientists, the problem is fundamental.
“LLMs are largely hopeless for robotics,” he says. Large language models like ChatGPT simply cannot process the messy, unpredictable data of the physical world. “They’re not a path towards human‑level or human‑like intelligence, or even animal‑like intelligence, because they cannot deal with real‑world data. They just are not built for that.”
At the VivaTech conference in Paris, LeCun explained that robots need AI systems capable of forming abstractions mathematical filters that strip away noise and leave only the useful structure of the world. In practical terms, this means an AI should understand that predicting the exact way a pen will fall is pointless; instead, it should grasp the broader physics of the situation.
That’s why Paris‑based AMI Labs is developing a completely new kind of AI, one not based on the architecture behind ChatGPT or its rivals. Investors are betting big on the shift: earlier this year, AMI Labs raised over $1 billion, backed by Nvidia and the private wealth fund of Jeff Bezos.
Their goal is to build AI that can operate flexibly in real environments the kind of intelligence robots need to safely handle objects, adapt to surprises and make decisions without relying on brittle, text‑trained models.
The robotics industry sees this as the next frontier: moving beyond chatbots and into machines that can genuinely understand the world around them. Whether AMI Labs or another challenger gets there first, the race to build real intelligence not just clever text prediction has officially begun.
+ There are no comments
Add yours