[jloughry]

Computer science. I worked for Lockheed for fifteen years---long enough to see some programming projects succeed and others fail. I wondered why the difference. I had access to all the project records, and Lockheed was cool with it, so I took copies of all the data and found an advisor willing to let me work on it.

The problem was deep and it ended up taking a lot longer than I expected to solve it, but that's the way of research. I did eventually discover some new things in the data, and got the PhD, and my thesis formed the basis of my new company.

You're right; I didn't get enough sleep. I got into the habit of working 13 to 15 hours a day. But I wasn't making progress on my thesis fast enough for the university's requirements, so I had to choose. I was still working 13--15 hours a day, seven days a week, but exclusively on my thesis. I finished in eight years. Without Lockheed, I would have been done in four.

Ironically, my day job for Lockheed was being the PI (Principal Investigator) for an Air Force research project that was essentially a whole other PhD thesis. [We took the Google Web 1T 5-Gram Corpus and inverted it; essentially we were doing adversarial ML (generators and discriminators) before that term had been invented.] That's why it took me eight years---because I was doing the work of two different full-time PhDs at the same time.

(Only one got me a neat robe to wear at the end, though.)