This is a constantly shifting field where the problems never get solved. As a result, this is still a good read, despite the submissions over time.

A previous discussion is here:

https://news.ycombinator.com/item?id=20455780

Very few of the points raised there have been adequately addressed.