2 ms·

If you are a junior engineer who just graduated, what would you do to ensure that you learn the needed skills and not be overly reliant on AI?

fragmede 2y ago
I wouldn't, because AI isn't the problem. With machines being able to run deepseek locally, the problem to look out for isn't the possibility that the web service will go down and you have to live without it, it's that, as their capabilities currently stand, they can't fix or do everything. I learned to program some time before AI became big, and back when I was an intern, I'd get stuck in a rut trying to debug some issue. When I got stuck in the rut, it would be tempting to give up and just change variables and "if" statements blindly, just hoping it would somehow magically fix things. Much like I see newer programmers get stuck when the LLM gets stuck. But see, that's where you earn your high paying SWE salary. For doing something that other people can not. So my advice to Jr programmers isn't to avoid using LLMs, it's to use them liberally until you find something or somewhere they're bad at, and look at that as the true challenge. With LLMs, programming easy shit got easy. If you're not running into problems with the LLM, switch it up and try a different language with more esoteric libraries and trickier bugs.
jcon321 2y ago
Well before AI, us old guys had to google our own issues and copy/paste from stackoverflow. When I was a junior I never thought about being overly reliant on "google or stackoverflow", but LLMs are slightly different. I guess it would be like if I googled something and always trusted the first result. Maybe for your question it means not copying/pasting immediately what the LLM gives you and have it explain. Wasting a few mins on asking the LLM to explain, for the sake of learning, still beats the amount of time I used to waste scanning google results.

fragmede 2y ago

I wouldn't, because AI isn't the problem. With machines being able to run deepseek locally, the problem to look out for isn't the possibility that the web service will go down and you have to live without it, it's that, as their capabilities currently stand, they can't fix or do everything. I learned to program some time before AI became big, and back when I was an intern, I'd get stuck in a rut trying to debug some issue. When I got stuck in the rut, it would be tempting to give up and just change variables and "if" statements blindly, just hoping it would somehow magically fix things. Much like I see newer programmers get stuck when the LLM gets stuck. But see, that's where you earn your high paying SWE salary. For doing something that other people can not. So my advice to Jr programmers isn't to avoid using LLMs, it's to use them liberally until you find something or somewhere they're bad at, and look at that as the true challenge. With LLMs, programming easy shit got easy. If you're not running into problems with the LLM, switch it up and try a different language with more esoteric libraries and trickier bugs.

jcon321 2y ago

Well before AI, us old guys had to google our own issues and copy/paste from stackoverflow. When I was a junior I never thought about being overly reliant on "google or stackoverflow", but LLMs are slightly different. I guess it would be like if I googled something and always trusted the first result. Maybe for your question it means not copying/pasting immediately what the LLM gives you and have it explain. Wasting a few mins on asking the LLM to explain, for the sake of learning, still beats the amount of time I used to waste scanning google results.