I have used Khan Academy in the past because I have been ploughing through books or articles on neural networks, and I often find I'm missing background - such as linear algebra, or statistics. Sometimes I sign in, many times I haven't signed in. But this is my idea. There is a certain background in a mixture of fields that is very useful for people who are interested in either the brain or in neural models. Statistics is one of those fields, calculus is another, linear algebra is a third. Recently I saw a video on the topic where 'statistical mechanics' would have been useful. Neural networks have taken a big leap in recent years, and now "deep learning" is used for all sorts of intelligent applications. One problem is that even researchers don't always know why the stuff works, just that it does (in some cases). So my idea is that Khan Academy offer a combination of existing courses that can bring people to competence in then taking on 'deep learning'. The courses do cover much of the basics, and if a  member who can learn the topic, you can put in any missing gaps in the courses you have. You could even add a "deep learning" course. It would not have to be a very practical one (the practical ones are full of boring tweaks to methods), but it could be an exciting one about new models and how they work. Some models have 'chaotic attractors' so that an input spirals down into the attractor basin, which is the solution to a problem, or a memory, others have a kind of bubbling up and down process as a percept gets more and more defined, and there are sorts of types of learning. So that's the idea for a future direction.