Very interesting thoughts about the fact that sometimes it pays more to add more data rather than fine-tuning the weights on your fancy machine-learning algorithm.

Second part of the very good article about having more data vs. better algorithms