Nope, that's still the wrong way round. If "usefulness" is Y and "knowledge" is X, and you needs lots of knowledge before it's useful, then the curve is shallow.

I see it something like: http://www.wolframalpha.com/input/?i=Plot[%28Sqrt[Pi]+Erf[3x...]. At first, you can use it and get something done, but it isn't useful. As you learn more, it becomes slightly more useful, but not much, until it hits a point where you finally understand the interface, and can use the features to their complete intent.

Having thought this through a little more, it seems more like bumps. For example, you can't do anything with Vim until you learn basic motion and insertion. That a bump. Then, you learn how to use the more complex motions, and get another bump. With something that has a steep learning curve it is done in bumps, rather than being smooth, because partial knowledge doesn't help any. If you think about it, if you only know the j and r commands in vim, you're not going to be doing anything at all, you're going to need to know at least hjkli before you can even use it.

I've decided that "steep learning curve" doesn't have to do with mathematical graphs at all. Saying something has a steep learning curve is the same as saying there is a barrier to entry. That barrier is knowledge, and a steep path over that barrier indicates a lot of of basic knowledge is necessary to get past the barrier.