The reason he says that a good startup idea looks like a bad idea is that good startup ideas must be hard, and if they weren't, many people would have already d

The reason he says that a good startup idea looks like a bad idea is that good startup ideas must be hard, and if they weren't, many people would have already done it by now. For example, if a startup were able to create a piece of software that could take a block of text and produce a summary (not keywords and such, but a real summary that a human would write if asked). I would think that that seems like (and IMO is) a good idea. Here, the problem is how do I do this rather than should I do this. The problem truly is, "How do I do this profitably?" Sure, lots of problems are hard. But there aren't very many hard problems that are also profitable. In my opinion, that's the greatest difficulty with a startup. It's not convincing investors that the problem is difficult. It's convincing investors that people are willing to pay for the solution. A historic example rather than a theoretical one could be Googles' basic concept of page-rank. It seems like a good idea (this is not just hindsight, IIRC there are even stories about Larry Paige explaining the concept at the time and people thinking that it was genius). Just, how many people at that time thought of it? Even if people did think of it, it still is a technically difficult algorithm when you take into account the sheer amount of data that it must process. So again, the problem is not should I, but rather, how do I. Again, you're misunderstanding the problem. Yes, PageRank was an "obviously good" algorithm. Heck pg himself thought of something very similar while he was at Yahoo. But you have to also remember that, at the time, web search was considered a "dead" field. Everyone thought that the future was manually created indexes and portals (like Yahoo) rather than in search. Google's real accomplishment was, IMHO, showing that web search could be profitable. Yes, they couldn't have done it without PageRank, but the new algorithm was only a necessary factor, not a sufficient one.