90% of the truth about ML is inconvenient
ML jobs are often full of wrong expectations. Have a look, what I think are some of the most common misunderstandings.
Hey guys! I once discussed with my past colleague that 90% of machine learning specialist work is, actually, engineering. That made me thinking, what other inconvenient or not obvious truths are there about our jobs? So I collected the ones that I experienced or have heard from the others. Some of them are my personal pain, some are just curious remarks. Don’t take it too serious though.
Maybe this post can help someone to get more insights about the field before diving into it. Or you can find yourself in some of the points, and maybe even write some more.
To get more content and make me feel better, please, subscribe and receive my new posts on email. I prepare all of them with love for you <3. Special warm thanks for the ones who share.
Anyways, have a good reading!
List of inconvenient truth about ML job:
90% of your job won’t be about training neural networks.
90% of ML specialists can’t answer (hard) statistical questions.
In 90% of cases, you will suffer from dirty and/or small datasets.
90% of model deployment is a pain in the ass. ( . •́ _ʖ •̀ .)
90% of success comes from the data rather than from the models.
For 90% of model training, you don’t need a lot of super-duper GPUs
There are 90% more men in Ml than women (at least what I see).
In 90% of cases, your models will fail on real data.
90% of specialists had no ML-related courses in their Universities. (When I was diving into deep learning, there were around 0 courses even online)
In large corporations, 90% of your time you will deal with a lot of security-related issues. (like try to use “pip install something” in some oil and gas company, hah)
In startups, 90% of your time you will debug models based on users' complaints.
In 90% of companies, there are no separate ML teams. But it’s getting better though.
90% of stakeholders will be skeptical about ML.
90% of your questions are already on StackOverflow (or on some Pytorch forum).
P.S. 90% of this note may not be true
Please, let me know if you want me to elaborate on this list - I can write more extensive stuff. And also feel free to add more or criticize some points.