I've just listened to the podcast and have a few questions. My background is only a bachelors in CS.
Let's assume no configuration.
1. Handling the variety of input in terms of syntax seems like it would be very, very troublesome. How?
2. The search space for appropriate models seems like this computation would take large amounts of time - on the order of days. How?
3. Outputting human readable text regarding meaningful inferences seems very challenging. I imagine that many conclusions would be, to a human, fairly obvious. How?
I do plan to look into their research but wouldn't mind discussion here.
