It's such a shame that they aren't releasing any detailed technical paper anymore on all the technical details of the model and how it was trained.

*Edit* Oh, I just read here (https://www.reddit.com/r/MachineLearning/comments/1armmng/d_...) that a technical paper should be released later today?