So everything that you see can be improved. What you want to find our improvements which change the market? Sometimes you can see this before you start working on the problem. I can only see the possibilities when I consider the problem deeply.
So my suggestion is to find a problem that you have, and figure out what is the least that you have to do to solve it. (You really want this to be 1 weeks work). Then develop the solution and figure out how it sucks, and it does, all software sucks. Get feedback from your friends and make improvements.
Once you get the urge to do your own thing an important stage is to reflect upon your ideas and their feasibility. Once you hit on the right one, you should get a "this is it" feeling. Seeing other sites doing the 'same' thing should just spur you on.
Two other bits of advice:
1. Find something limited in scope, and build it under constraints. At least to start, assuming that you're going to keep your day job. In other words: can you build something that actually has value, and that you would use, in 40 hours? This is a good exercise for a first startup project, and can keep you from running down a rabbit hole.
2. Decide if this is a side project or a full-time job. If it is a side project, you can build something you like and help it grow slowly, without worrying about your take-home pay right away. If you're going to quit your day job, the stakes are much higher - you can't rely on slow growth (because your rent/mortgage depends on profit), and you should never rely on explosive viral growth (because it probably won't happen to you, no matter how good you are).
I would suggest making something that exists, better. Something that people will want and will use because they are already trained to use it.
Go one step further and make sure that it runs on people's system tray as well as on the web, and on a mac, and linux.
One way might be to combine a platform (PC/Mobile/Voice) with a communication medium (RSS/Twitter/Video) and think about ways people could use a new combination. For example, Voice + RSS = people 'reading' the web from listening to a phone.
Another way might be if you are using a service already but there is a serious 'feature gap' that you could address, i.e. how to find out what Twitter users are linking to? Sure you would have to de-code tinyurl addresses, what else would you have to do to accurately report that figure? (Idea was suggested by Merlin Mann; he is another guy who frequently posts ideas he'd like to see).
In YouTube's case, I argue that they knew how to market themselves. Don't discredit your positioning and target audience. Write a kickass app tailored to what they specifically need. And hey, if it's another video app, how is your version better or more niche than YouTube?
If you can do it better, you can succeed. (though don't try, for example, to create a mail app like Gmail, you will probably never keep up with them!)
And in your case, "Build Something You Want" (provided it doesn't exist) - and you'll have more energy to play with too.
Facebook wasnt the first social network, but they did it. Basecamp wasnt the first project management tool. They took something that they thought wasnt great and made it better.
"Idea guys" as mentioned earlier often come to you with a "world changing" idea. Offer you 50% of the company and you, the tech guy assume all the time and risk. If you do partner with someone, find someone who brings a viable skill to the table and not just a pile of ideas.