Why have a "standard route" at all? It seems that the greatest poxes facing the ML community are intellectual homogeneity & group-think. Having everyone take too-similar paths to PhD appears (to me) the very pathology implicated. Granted, as a strange-path freak, I'm biased. https://twitter.com/tallinzen/status/1264941450099331072

A finer-grained dig: I'm most bugged by the idea that everyone’s outside exp. shd be SDE work. Certainly those skills are fantastically useful (we should have some SDEs!), but the intellectual terrain is limited and I’d hate for a whole PhD class to bring only engineering skills.

Final thought, this is a strong disagree on a remark that I believe was intended in the spirit of soliciting strong disagreements from various angles. No disrespect intended towards my mate @tallinzen (& suspect that we actually agree on the aspect I pushed in this thread).