There is a huge value in on-the-job training, but that value comes from a direct alignment between the work and the learning.
Time delays in the University model (for one thing) make it hard to produce what is needed. For instance if you want to "learn deep learning", you could get a PhD, and maybe 9 years later you will be an expert, but 9 years is a very long time and the field may be tapped out and deep learning PhD's will be driving cabs.
Time delays in the University model (for one thing) make it hard to produce what is needed. For instance if you want to "learn deep learning", you could get a PhD, and maybe 9 years later you will be an expert, but 9 years is a very long time and the field may be tapped out and deep learning PhD's will be driving cabs.