In fact the usual advice is that you shouldn't be doing something technically virtuosic in a startup. If your startup's critical path includes writing something as technically impressive as the Glasgow Haskell Compiler, you might have trouble...
Though there are a few people around here who've built software businesses around things that are also nontrivial technical advances; cperciva comes to mind.
I know that's a sentiment that's thrown around here, but I think it's simplistic.
I live in Israel, where a very common path to an Exit is to build some sophisticated technology, then sell it after a few years to a larger company. So it is a very possible path you can take. It has its own advantages and disadvantages over the much more visible, much more HN-friendly, "Consumer Internet Startup".
The exit and valuation models for these two approaches are also different. In the former, some company buys you because they want to use your technology in their own products, in the latter, some company buys you to get your customers/users.
It seems all or most pg's articles just say the opposite: a startup should climb the highest hills, and can do so faster the big corporations because they are light-weight. If I'm not mistaking, "climbing" here do not mean find a genius idea or a disrupting design for a search field, it means developping something hard in software.
The conclusion should be that a startups that is not doing something technically hard should be considered in trouble (because so easy to clone).
I think it does not have to mean "developing something hard in software", but more like "finding an unsolved problem and using software to solve it." It does not have to be hard to solve in the software side. I guess the hardest part is finding the right problem to solve!
What I think, for what you may think it's worth, is that you should climb the highest hills, that you'll have to climb, right away. Best to know you can do it before wasting time.
But, that you should look for the easiest way to success. It'll take months for a large company to have the meetings to acknowledge your niche, let alone try to serve it. The only thing you have to fear is another startup and you're at the advantage is that situation.
Building huge walls or picking a niche with a large cost to serve, is like wasting time on DRM for your game, instead of working on marketing and having someone actually buy it.
The usual advice seems odd to me. There are two extremes: you can invent something technically trivial and take the market risk, or you can invent something that's technically challenging but with obvious marketability.
Though there are a few people around here who've built software businesses around things that are also nontrivial technical advances; cperciva comes to mind.