In the age of artificial intelligence, renowned tech labs, including Anthropic, OpenAI, and Meta are talking about reaching superintelligence by scaling the large language models (LLMs).
However, the Godfather of AI and Ex-Meta AI chief, Yann LeCun, holds a contrary view, calling LLMs like ChatGPT-4 a fundamental dead end for achieving human-level intelligence.
Recently, the pioneer of modern AI has come up with a revolutionary vision: world model AI systems, a new breed that can understand the real physical world with real intelligence.
To pursue his groundbreaking ambitions, Yann LeCun has raised $1.03 billion, as the AI chief seeks to commercialize artificial intelligence systems built around reasoning, planning and "world models."
The funding round was also co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital and Bezos Expeditions.
The “playbook” used by almost every major AI company today relies on autoregressive LLMs. LeCun believes this approach is limited because it lacks grounding in reality.
The founder of Advanced Machine Intelligence (AMI) has bluntly spoken about the limitations of AI chatbots, such as Claude, Gemini and ChatGPT. But he never dismissed the overall utility of LLMs.
According to LeCun and reported by Wired, “It's true that [LLMs] are becoming really good at generating code, and it's true that they are probably going to become even more useful in a wide area of applications where code generation can help.”
“That’s a lot of applications, but it’s not going to lead to human-level intelligence at all,” he added.
LeCun also argued that LLMs only predict the most likely next word, they don't actually "know" facts. They are just mimicking the structure of human language.
Moreover, another limitation of LLMs lies in the lack of persistence and memory in these large language models. They do not possess long-term memory of interaction; they just operate within a fixed context window without learning any new experiences.
On the other hand, to develop human-like intelligence, a persistent memory of past events and new experiences are always required.
LeCun also claims that real intelligence does not begin with words; it starts with action. In today’s world, generative AI has mastered the art of predicting language. But, the physical world is far too chaotic for simple text-based architectures.
According to the company’s statement, “AMI aims to build a new breed of AI systems that understand the world, have persistent memory, can reason and plan, and are controllable and safe.”
“Our main goal is to build intelligent systems that understand the real world,” the company added.
AMI aims to build a new frontier dominated by “action-coordinated world models” that can evade noise of unpredictable sensor data to focus on what matters the most: the representation space.
These models will also allow agentic AI systems to process and understand the consequences of their actions, thereby bringing reliability and safety to the treacherous AI landscape.
LeCun says AMI seeks to work with companies in manufacturing, biomedical, robotics, and other industries that have lots of data.
AMI will also develop a “universal world model,” providing the basis for a general intelligent system. He does not mention a definite timeline regarding the release of the first world model.
Yann LeCun left Meta in November 2025 and founded the AMI startup to pursue an alternative AI approach.