A new study from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) demonstrates a significant advancement in AI’s ability to understand and reason about the physical world. Researchers developed a framework that allows large language models to create and utilize internal, simplified simulations of objects and their interactions, moving beyond pattern recognition to more intuitive, physics-based reasoning. This approach improved the models’ performance on tasks requiring an understanding of object dynamics, such as predicting how a stack of blocks might fall. The work suggests a pathway toward AI systems that can learn and reason with less data by building internal world models, a step closer to more general and common-sense artificial intelligence. Read the full article at https://www.technologyreview.com/2024/05/15/1092631/ai-models-can-excel-at-reasoning-with-a-bit-of-help-from-physics-simulators/