Yann LeCun, Meta's chief AI scientist, plans to leave the company to start his own venture focused on world models. His departure comes as Meta restructures its AI division amid increasing competition from rivals. LeCun has expressed skepticism about the current state of AI technology.

Yann LeCun, a key figure in AI development, is increasingly at odds with Meta's direction. He is considering leaving the company to start a venture focused on world models, which he believes could advance AI more effectively than Meta's emphasis on language models.

Yann LeCun, Meta's chief AI scientist, is exploring the possibility of leaving the company to create a startup focused on world models. He has been in talks with associates and investors about this new direction, which diverges from Meta's current focus on large language models. His plans remain uncertain, and he may decide to stay with Meta.

This article presents a codebase for a study on how unified multimodal models (UMMs) enhance reasoning by integrating visual generation. The research introduces a new evaluation suite, VisWorld-Eval, which assesses multimodal reasoning capabilities across various tasks. Experiments show that interleaved visual-verbal reasoning outperforms purely verbal methods in specific contexts.

Yann LeCun, Meta's chief AI scientist, plans to leave the company to launch his own startup focused on world models. His departure comes as Meta restructures its AI unit in response to competition from rivals like OpenAI and Google. LeCun has expressed skepticism about the current hype surrounding AI technologies.

Runway, a video AI company, raised $315 million, boosting its valuation to $5.3 billion. The funding will support the development of world models, advanced AI systems that can predict real-world scenarios, enhancing applications in robotics and self-driving technology.

The article examines how AI models, particularly LLMs, struggle with adversarial reasoning compared to human experts. It highlights the importance of simulating interactions and anticipating responses in competitive environments, contrasting this with the limitations of current AI in understanding the depth of human decision-making.

The article argues that the key barrier to developing Physical AGI is the lack of diverse and abundant data compared to human experiences. It emphasizes the need to capture human sensorimotor experiences through egocentric video to train models that understand and predict physical interactions. The author believes this approach can bridge the gap between human knowledge and robotic capabilities.

The article explores the growing interest in world models across major AI labs, detailing their potential to simulate environments and predict outcomes. It contrasts these models with current AI systems, emphasizing their ability to manage complex, adversarial domains through a feedback loop that enhances learning over time.

This article discusses the progression of video generation techniques towards creating comprehensive world models that simulate real-world dynamics. It outlines a four-generation taxonomy, highlighting how each generation enhances capabilities like realism, interaction, planning, and stochasticity. The authors emphasize the importance of integrating physical and mental world models for applications in robotics and AI.

This article introduces Reinforcement World Model Learning (RWML), a method that helps large language models (LLMs) better predict the outcomes of their actions in various environments. By using self-supervised learning to align simulated and actual states, RWML improves the agents' ability to adapt and succeed in tasks without requiring external rewards. The authors demonstrate significant performance gains on benchmark tasks compared to traditional approaches.

The article outlines four emerging AI research trends crucial for enterprises: continual learning, world models, orchestration, and refinement. These trends focus on enhancing AI applications by improving memory retention, simulating real-world environments, optimizing resource use, and enabling self-improvement processes.

The publication introduces CWM, an open weights large language model designed to facilitate research in code generation utilizing world models. It aims to enhance the understanding and application of code generation techniques in various domains. The model is made available for researchers to explore its capabilities and contributions to the field.

The neural motion simulator (MoSim) is introduced as a world model that enhances reinforcement learning by accurately predicting the future physical state of an embodied system based on current observations and actions. It enables efficient skill acquisition and facilitates zero-shot learning, allowing for a decoupling of physical environment modeling from the development of RL algorithms, thus improving sample efficiency and generalization.

The essay critiques various perspectives on world models, which are essential for developing virtual agents with artificial general intelligence. Drawing from sci-fi and psychology, it emphasizes that a world model should simulate all actionable possibilities of the real world for effective reasoning and action, and proposes a new hierarchical architecture for such models within a Physical, Agentic, and Nested (PAN) AGI framework.

A new benchmark for generative world models (WMs) is introduced, focusing on their effectiveness in closed-loop environments that reflect real agent-environment interactions. This research emphasizes task success over visual quality and reveals that controllability and effective post-training data scaling are crucial for improving embodied agents' performance. The study establishes a systematic evaluation framework for future research in generative world models.

Genie 3 is a groundbreaking world model developed by Google DeepMind that generates interactive environments in real-time, allowing users to navigate and interact with them based on text prompts. It enhances previous models by improving consistency and realism, supporting complex simulations and interactions while paving the way toward advancing artificial general intelligence (AGI). The model also faces limitations such as a constrained action space and challenges in accurately representing real-world locations.

Meta has unveiled its new AI model, V-JEPA 2, designed to enhance understanding of 3D environments and physical object movements, enabling more human-like decision-making. This open-source world model aims to improve technologies like delivery robots and self-driving cars by allowing machines to reason about their surroundings without extensive labeled data. CEO Mark Zuckerberg's focus on AI is underscored by a planned $14 billion investment in artificial intelligence firm Scale AI.

Elon Musk's xAI is entering the competition to develop world models, aiming to create advanced AI systems capable of understanding and designing physical environments. The startup has recently recruited specialists from Nvidia to enhance its efforts in leveraging videos and robotic data for training these next-generation AI models.

The article discusses the limitations of large language models (LLMs) in relation to understanding and representing the world as true models. It argues that while LLMs can generate text that appears knowledgeable, they lack the genuine comprehension and internal modeling of reality that is necessary for deeper understanding. Furthermore, it contrasts LLMs with more robust cognitive frameworks that incorporate real-world knowledge and reasoning.

MindJourney is a new research framework that enables AI agents to explore simulated 3D environments, improving their spatial interpretation capabilities. By using a world model and a spatial beam search algorithm, MindJourney allows AI to generate multiple perspectives of a scene, enhancing its ability to answer spatial questions without additional training. This approach significantly boosts the performance of vision-language models, suggesting potential applications in robotics and smart technologies.