For senior-level roles (sorry, tier 3 data analyst making pie charts in Powerpoint for no-name business probably won't get paid mid-six figures, and therefore isn't under personal consideration.) *: or other research-heavy areas - e.g. AI and deep learning
Someone will make a framework that every Tom dick and Harry in software can learn and we will be hiring kids out of high schools.
Yeah, that's annoying when people apply models and draw spurious results without even checking the assumptions (e.g. inference following linear regression). Practitioners should at least have the sense to acquire an intuitive understanding of the inner workings. To be clear, I believe that shouldn't require rigorous functional analysis for SVM or algebraic-topological data analysis for classifying manifolds. Proper methodology doesn't always necessitate proofs.
Most people who don’t have data science experience do shitty ML or draw terrible conclusions. Fine as ML infra engineers though
Currently, need PhD for most senior/exec roles. Will be interesting to see if changes over time as more bachelors learn on the job
I think it is very, very hard to become an L7+ IC AS at Amazon without a PhD, or an L6+ RS at Google. Better off moving to manager track at senior level.
Does this apply only to ML positions?
The post was about ML roles in big tech, which I took to mean applied scientist/research scientist IC roles. For SDEs, lack of a PhD does not present a glass ceiling issue.
I find that most people learning on the job just learn hacky ways to do things, but don't develop enough breadth of knowledge about various techniques to be able to always choose the right tool for the right job and also lack the depth of knowledge to know how to correctly apply more complex theory in a practical manner. Some tiny amount of people are geniuses that can teach themselves this outside of an academic environment. Most people can't.
agree. i feel this applies to not only ML roles but also other software engineering roles
Edit: "with at least" -> "generally need at least"