I won't claim to know what he thinks exactly -- maybe he's the alum you refer to? -- but I wonder if Samuel Clay (Newsblur, YC S12) would agree with that sentiment, given that he has published https://github.com/samuelclay/NewsBlur for all to see. I haven't committed to my own startup idea yet, but my plan for it does involve open-sourcing the code while keeping a custom corpus and its associated NLP models private.