Not really, not in ML. You can't do anything in ML without writing some kind of program, even if it just glues together sklearn or pytorch calls. I'm not talking about advanced algorithm knowledge, but merely being able to cobble together the kind of program their CV implies they do on a daily basis.

Again, I don't know if it applies to the Reddit poster, but I sure met people like this.

One example, not an academic, but someone who claimed he "revolutionized" his Important Department at Big Corp using ML and NNs, and didn't know what gradient descent is. I don't mean he couldn't implement it on the spot, rather be never heard of it, and couldn't understand when I described it to him.

Another one, PhD candidate in computer science at top-10 uni, self-described "expert" in C++, could not, indeed, write a C++ for loop over an integer range. Or indeed a for loop in any language of his choice - he had known, but forgot. He had a number of publications that sounded sane.

Anyway, if someone is an ML academic, wants an industry job, and can't code enough, that would probably lead to the described outcome. And again, I'm not jumping to that conclusion.

It seems to me like you've just described hiring a mathematician on arithmetic skill. If you hire this way of course many will use code practice tools to pretend they are juniors or incapable of delegating since most of their time should be occupied with less low level ways of doing things or assigning things.

I can go a long time without writing for loops, the better I know the surrounding tools in an ecosystem, the less I accidentally reinvent something with details that low.

I don't remember all that much about pandas, but enough to try to use it to do a matrix operation instead of writing the for loop if I were assigned a ML task for some reason.