Keep pulling the thread on Yann LeCun.
The Joint-Embedding Predictive Architecture (JEPA) is designed to find a trade-off between maximizing the information content of its representation while only preserving information that is predictable.
In the Joint-Embedding Predictive Architecture (JEPA), the level of detail in the learned representation is determined by the time horizon used for training the prediction.
A "mental world model" is a type of AI model designed to predict human behavior and mental states.
An effective educational dialogue system would require a 'mental world model' to predict a student's mental state and their ability to absorb new information.
The proportion of women in U.S. computer science undergraduate programs has declined from approximately 40% in the 1980s to a current national average of 15-20%.
The proportion of women in computer science PhD programs is roughly the same as in undergraduate programs, indicating no significant "leak in the pipeline" at the graduate level.