Isn't there a way simpler timing analysis. The blog post is about a conference paper which will be presented in CVPR and the conference starts in 2 days.
I think if the problem is figuring out an ML solution to an already existing large dataset (as in the case of Kaggle and big companies), you do not need an advanced education. However, if you are doing a start-up one needs to answer questions like when should I stop collecting data, when should I give up trying this algorithm, when should I conclude this problem is impossible in its current form etc. These questions require crazy amount of experience of solving novel problems. During your PhD, you try to solve many novel problems and it makes you an expert to answer these questions. I also think you need an advisor/mentor to develop these skills. This is the real knowledge you can learn in PhD and that's why they are valuable and hard to replace