This document discusses building better models in cognitive neuroscience. It outlines several key ideals for models, including being consistent with known neuroscience and keeping models as simple as possible while still capturing important behaviors and learning mechanisms. The document then covers different approaches to modeling units, connections, and learning in neural networks. Specific models discussed include leaky integrate-and-fire models and the Izhikevich model for units, as well as long-term potentiation and depression for learning. The document concludes by discussing how to model behavior and decision making using neural network outputs and competition between choices.