Test with at most 5 users

There's an interesting post from back in 2000 on Jacob Nielsen's site here. He makes a good case for why you don't need to do usability testing with any more than 5 users. This graph is illustrative, but shows how the learning from usability testing has diminishing returns.

I spoke with John Briggs, the first product manager at Yahoo, and he had mentioned that you only need to be 70% right before you ship (and iterate in the field to get to 100%). He also mentioned that you can learn everything you need to learn to get to 70% by talking to 3 users. So 2 independent expert sources are saying almost the same thing here.