Machine learning researchers build a better world model

Machine learning researchers, including Randall Balestriero and Yann LeCun, have developed a counterfactual world model technique designed to improve how physical AI systems like robots navigate the real world. The approach enables zero-shot physics inference by using the mismatch between model predictions and reality to calculate alternative scenarios with a single subtraction. According to the researchers, this helps robots recover from unexpected events, adapt to faults, and debug failures more effectively. Open article →