Interestingly, the common use of this term is completely, unutterably, WRONG! Wherever I look (particularly in rigourous, exact disciplines such as computing, but also one of the softlinks below does this) I see "steep learning curve" meaning hard to learn.
Really? As explained by Agthorr above, the learning curve is a (hypothetical, mythical, nonexistent) graph of how much one knows of something after spending X effort+time. Steep curves are the one which go up faster (we assume the learning curve is the graph of a monotone function here...). Would you rather (assuming -- incorrectly -- that you could do both and end up with the same amount of knowledge) Teach Yourself C++ In 21 Hours or Teach Yourself C++ In 21 Years?
Thought so. Yet many insist on claiming that the necessarily steeper learning curve of Teach Yourself C++ in 4 Hours is worse!
A probable source for this confusion is the intuitive expectation that steep ascents are "harder". This is true -- trying to learn C++ in 7 days wastes time rather than saving it, as you waste the first 7 days "learning" C++, the next 28 days discovering that you need to learn in properly, and only then can start studying C++. But all this shows is that the "2.4 days strategy" is ineffective. And the reason it is ineffective is that it gives a very shallow learning curve: you spend the first 14 days learning nothing.
Learning curves typically also factor in effort. So there really is no excuse.