| 1. I have pretty much a pure programming background. No previous experience with speech or machine learning when I was hired (but that was long ago). 2. The software engineers in the team have a variety of backgrounds. I think I may be the only PhD, and my subject was not relevant to the job. Plenty of people with bachelors (for visa reasons, non-US employees tend to have higher degrees). Machine learning knowledge helps, but is not strictly required. For the data scientists, on the other hand, advanced degrees are a definite plus, and so is some specialization in a relevant subject. 3. The most important thing to know about Siri's internal working is that we don't talk about Siri's internal working… If you randomly would like to learn something to improve your chances of working at Siri, a machine learning class (e.g. Ng's and/or Hinton's Coursera classes) could definitely help. 4. There is still a lot of engineering involved, and often by the time a formal schedule is worked out, the scientific discovery part has largely been solved. Sometimes features end up not working out and have to be pushed back. What helps for Siri is that a lot of the complex functionality is server side or in updatable assets, so iteration is a possibility. 5. That's above my pay grade, really. What I learned the past few years is never to bet against deep learning being able to tackle a particular problem, but I can't shake the feeling that we'll discover limits some day. |