Yann LeCun’s new venture is a contrarian bet against large language models

The Future of AI: A Conversation with Yann LeCun

Yann LeCun, a renowned AI researcher and Turing Award recipient, has long been a contrarian figure in the tech world. He believes that the industry's current obsession with large language models (LLMs) is misguided and will ultimately fail to solve many pressing problems. Instead, he advocates for world models – a different type of AI that accurately reflects the dynamics of the real world. LeCun is also a staunch supporter of open-source AI and criticizes the closed approach of frontier labs like OpenAI and Anthropic.

A New Venture: Advanced Machine Intelligence (AMI)

LeCun recently announced the launch of Advanced Machine Intelligence (AMI), a global company headquartered in Paris. AMI aims to provide a credible alternative to the US-China binary, offering a more open and inclusive approach to AI development. LeCun sees this as an opportunity to create a platform that is neither Chinese nor American, but rather a global effort that prioritizes sovereignty and control.

The Problem with LLMs

LeCun is vocal in his criticism of LLMs, arguing that they are overhyped and limited in their capabilities. He believes that LLMs can only manipulate language well, but are unable to truly reason or plan, as they lack a model of the world. LeCun argues that the truly difficult part of AI is understanding the real world, and that LLMs are not equipped to handle this task.

World Models: The Future of AI

LeCun's solution is to focus on world models, which learn to represent videos and make predictions in an abstract space, ignoring the details that cannot be predicted. This approach is based on the JEPA (Joint Embedding Predictive Architecture) framework, which trains AI models to understand the world by learning from observation. LeCun believes that world models have the potential to revolutionize AI, enabling systems to reason and plan in the real world.

Applications and Implications

The applications of world models are vast, with potential uses in complex industrial processes, smart glasses, and agentic systems. LeCun envisions a future where AI systems can assist humans in a more meaningful way, enabling them to take actions in the world without making mistakes. He also sees the potential for world models to unlock Level 5 autonomous driving and truly useful domestic robots.

The Role of Academia

LeCun believes that academia should focus on long-term objectives that go beyond the capabilities of current systems. He argues that academia should invent new techniques and focus on the next big thing, rather than refining the last one. LeCun sees the University of Montreal as a hub for innovation, where researchers are working on breakthroughs that will shape the future of AI.

AMI Labs: A New Era in AI Research

LeCun's new company, AMI Labs, will be headquartered in Paris, with offices in North America and Asia. He plans to recruit top talent from around the world, including researchers from OpenAI, Google DeepMind, and xAI. LeCun sees AMI Labs as a place where researchers can work on world models and other cutting-edge AI projects, without the constraints of a traditional research lab.

Conclusion

Yann LeCun's vision for the future of AI is one of open-source, inclusive, and world-model-based development. He believes that LLMs are overhyped and limited, and that world models have the potential to revolutionize AI. LeCun's new company, AMI Labs, will be a hub for innovation, where researchers can work on world models and other cutting-edge AI projects. As the AI landscape continues to evolve, LeCun's vision will be worth watching.

Deep Dive

Artificial intelligence is a rapidly evolving field, with new breakthroughs and innovations emerging every year. In recent years, we have seen a growing trend towards the development of large language models (LLMs), which have the potential to revolutionize the way we interact with technology. However, LLMs are not without their limitations, and many experts believe that they are overhyped and limited in their capabilities.

One of the key challenges facing LLMs is their lack of understanding of the real world. While they can manipulate language well, they are unable to truly reason or plan, as they lack a model of the world. This is where world models come in, which learn to represent videos and make predictions in an abstract space, ignoring the details that cannot be predicted.

World models have the potential to revolutionize AI, enabling systems to reason and plan in the real world. They can be used in a wide range of applications, from complex industrial processes to smart glasses and agentic systems. In the future, we can expect to see world models being used in even more innovative ways, such as in Level 5 autonomous driving and truly useful domestic robots.

As the AI landscape continues to evolve, it is likely that we will see a growing trend towards the development of world models. This is an exciting time for AI researchers and developers, as we have the potential to create systems that can truly assist humans in a meaningful way. However, it is also a challenging time, as we must navigate the complexities of developing world models that can reason and plan in the real world.

What's Next for AI in 2026

As we look to the future, there are several trends that are likely to shape the development of AI in 2026. One of the key trends is the growing use of world models, which have the potential to revolutionize AI. We can also expect to see a growing trend towards the development of more inclusive and diverse AI systems, which can be used in a wide range of applications.

Another trend that is likely to shape the development of AI in 2026 is the growing use of explainability and transparency in AI systems. This is an area that is becoming increasingly important, as we need to be able to understand how AI systems are making decisions and why they are behaving in certain ways.

In addition to these trends, we can also expect to see a growing trend towards the development of more autonomous and self-aware AI systems. This is an area that is becoming increasingly important, as we need to be able to create systems that can truly assist humans in a meaningful way.

Overall, the future of AI is likely to be shaped by a number of trends, including the growing use of world models, the development of more inclusive and diverse AI systems, the use of explainability and transparency in AI systems, and the development of more autonomous and self-aware AI systems. As we look to the future, it isges that we will see a growing trend towards the development of AI systems that can truly assist humans in a meaningful way.

Meet the New Biologists Treating LLMs like Aliens

In recent years, we have seen a growing trend towards the development of large language models (LLMs), which have the potential to revolutionize the way we interact with technology. However, LLMs are not without their limitations, and many experts believe that they are overhyped and limited in their capabilities.

An AI Model Trained on Prison Phone Calls Now Looks for Planned Crimes in Those Calls

In recent years, we have seen a growing trend towards the development of AI models that can analyze large amounts of data and identify patterns. One such model has been trained on prison phone calls and is now being used to identify planned crimes.

The model uses a combination of natural language processing and machine learning algorithms to analyze the content of the phone calls and identify potential indicators of planned crimes. This includes looking for keywords and phrases that are commonly associated with criminal activity, as well as analyzing the tone and language used by the individuals making the calls.

The model has been shown to be highly effective in identifying planned crimes, with a high degree of accuracy. This has the potential to be a valuable tool for law enforcement agencies, who can use it to identify and prevent crimes before they occur.

However, the use of AI models to analyze prison phone calls also raises concerns about privacy and civil liberties. The model is trained on data that is collected from individuals who are incarcerated, and it is not clear how the data is being used or who has access to it.

As the use of AI models in law enforcement continues to grow, it is likely that we will see more and more concerns about privacy and civil liberties. It is essential that we have robust safeguards in place to protect individuals' rights and ensure that AI models are used in a way that is transparent and accountable.

Stay Connected

Get the latest updates from MIT Technology Review, including top stories, upcoming events, and more. Enter your email address to stay connected and receive special offers.

Source: https://www.technologyreview.com/2026/01/22/1131661/yann-lecuns-new-venture-ami-labs/

Yann LeCun's new venture is a contrarian bet against large language models