AI godfather warns language models are a 'dead end,' says world models are the future

Yann LeCun’s departure from Meta highlights a critical ideological rift in artificial intelligence between rapid commercialization and long-term scientific rigor. This divergence underscores the limitations of current large language models, which rely heavily on text data. Without embodied sensory experience, these systems struggle to understand the physical world, plan complex actions, or perform causal reasoning at a human level, revealing a significant gap between existing technology and true general intelligence. LeCun is now founding a startup focused on “world models,” an approach that integrates visual and spatial data to help AI predict outcomes based on physical dynamics. This paradigm shift suggests that future AI must learn from real-world interactions rather than static text corpora alone. By prioritizing spatial intelligence, this new direction aims to build systems that can reason about cause and effect, offering a more robust foundation for artificial cognition compared to current text-centric architectures. This development is highly relevant to open data because it necessitates the creation and sharing of new types of datasets. Moving beyond language requires large-scale, publicly accessible collections of embodied data, including video, spatial information, and sensor logs. As the industry debates the future of AI, the success of world models depends on open, diverse datasets that allow researchers to train systems on real-world physical interactions, challenging the dominance of proprietary, text-heavy data sources.

Source: ynetnews.com
Published on 2025-11-18

AI godfather warns language models are a 'dead end,' says world models are the future

Related news