ML too easy?
has ML become too easy for anyone to do these days? companies are hiring kids with bachelors degrees and barely a few years of work, sometimes not even.. I feel like getting a PhD wasnt that useful in getting a huge leg up .. 🤷🏻♂️
It's not too easy for complicated problems.
Easy to just use a package, but most of these BA kids don't know the underlying math and don't appreciate the assumptions or the strengths and shortcomings of each model. Also, they don't know what to do if they have to take one step beyond the using canned code in python/r
How to start ML?
I like an analogy of Dave - co-author of Pragmatic Programmer. High level car knowledge can help you to run a car, but if things break down - good mechanician is needed.
I was at the Tableau conference 2 weeks ago and the automation tools in the data science space is pretty amazing.. but that still was at a high level. I still think today only BA level in ML is super restrictive. You can’t even enter many research or interesting/unique jobs without at least a masters. These canned tools also yield generic results, they become useless in anything specialized. Aka where innovation happens
For basic things, sure. But all tech is like that. Kids aren’t going to be training models to make great meal recommendations any times soon.
neither is the guy trying to justify his phd.
I think there's has been lots of interesting research that's not fully applied yet, phd is really good if you want to do research, if you want to simply implement and improve existing products, I don't see why you would need that. significant percentage of our time spent "engineering", a small percentage is spent modeling.
Knowing what works when is where you can add value. ML won’t magically solve business problems.
Math!! 👍🏼
Where did you see math in that remark?