Two engineers debate whether you need to learn machine learning fundamentals when APIs and tools do the heavy lifting. They agree it depends on your job: research and model development require it, but some production roles only need to know how to call an API.
Chapters
Foreign.
It's really very easy to use these days compared with a few years ago when we have to learn actually the nuts and bolts of how to build RNPs and CNNs like neural nets and UNET, all those things. I felt like it's a, it's history now. And I'm curious in your opinion for new graduates or engineers who are not in the AI space and who haven't worked with these neural network models and who want to be part of the landscape, they want to maybe look for jobs related to AI engineers. Do you think it's still necessary to learn the basics of how CNN work, RNN's, neural nets in general, or language model fine tuning on those basics?
It depends. Learning the basics is always necessary in my opinion. You may not need them at the beginning or at your job. You don't use RNN nowadays compared to four years ago. They're still in use, but in many cases you may not use them at the moment. But I think the more you understand and learn about these things, fundamentally it's going to help you down the road much better. Yes. If you just want to be working as so called AI engineer, you know, using which is more of a software engineer with some extra skills, a little bit of knowing about machine learning, how to use these APIs and all that.
I don't even think you need to know machine learning. I don't think you need to know like classification models anymore because yeah, again.
It depends on the job. If you, if what you do is simply just write some code, which is more of a software engineering than really machine learning, you just need to write some code to integrate different APIs and things like that, which is pretty common nowadays, right? We are doing it even for different products and all that. So this is something that, yes, you don't need to go and learn about logistic regression or more traditional approaches. However, that said, I think in many cases when you are really going to solve a problem, then if you know about all of those approaches, traditional machine learning approaches and everything, when you want to implement a solution because you know them, then you know which approach is probably a better fit for your use case. But if you don't know that such a solution exists, you just always, your mind goes towards let me just use an LLM or use this ready made approach and things like that. That's why I think for students, for people who are more into graduate masters and Ph.D. that kind of thing, of course they need to learn these things. But if you just simply want to explore these capabilities and machine learning models and all that. You do not need to know all the fundamentals and probably theories, a lot of the math, linear algebra, things like that related to these things because you just simply want to call a bunch of APIs, right? But if you are an engineer or part of a machine learning team where you need to develop fine tune to do more fundamental things with these models, then I think you should know the basics. So that's what I think. That's my personal opinion.
