we are not coding, only using frameworks (easy) to implement customer requirements under constraints and this is hard part. AI will not make decisions what is better.
What a nonsense article. I'm tempted to just reply "This is your brain on LLMs" but I'll try to be less snarky.
Imagine having a toolbox, with different types of tools, and not knowing anything about them. Not knowing when it's appropriate to use one type of hammer over another.
That's the outcome of this, "Let LLMs take the wheel" attitude.
> Imagine having a toolbox, with different types of tools, and not knowing anything about them. Not knowing when it's appropriate to use one type of hammer over another.
Is this pro-LLM, or anti-LLM?
I agree with the truth of the statement, but it works both ways: the world is a huge toolbox, even just code is, even just iOS is.
There's far more to learn than I can even enumerate, so much so that I can't even tell you what percentage are unknown to me, 90%, 99%, or more.
LLMs are jacks of all trades and masters of none. I'm a matter of iOS coding, but I know less than jack about, say, metalwork (where I did one small project in secondary school).
Sometimes "jack of all trades" is useful. But always check their working, because it's a miracle they can ever tell fact from fiction and were definitely trained on both.
And when it comes to hammers, the machine probably does know more than me, as I can basically only functionally distinguish them by size and if they're rubber or steel, I can't say why some have wedges on the side that doesn't strike a nail while others have balls.
What happens in 5-10 years when new "engineers" come into the workforce who do not know that what the LLM has told them is incorrect?
I use LLM tools all day but because I used to do this all by hand I know what is wrong and what isn't. A graphing calculator is never wrong, all these LLMs make a lot of mistakes.
Someone just posted on a car forum what Gemini AI told them the required torque for a wheel bolt on a specific car is supposed to be. It was off by a huge factor to what is stated in the manual which would result in the wheel falling off at some point.
I have to disagree here. AI tools are great and work great. Until they do not. The speed with which they spit out a result gives a great sense of confidence and you may overlook an edge case which they do not cover. And it's not a question of if but when you will find out that you've crashed a million systems worldwide or made the wrong update on a 10TB table which ran for 3 days. I almost did the latter just yesterday because I was lazy to write a seemingly dumb and trivial function. Luckily I am pedantic enough to quadruple-check before I hit enter.
Leetcode and coding competition problems were never once thought to be a good representation of our ability or skill in the Software Engineering area.
It just became really important in the last decade or so with Big Tech using them to filter candidates as a proxy for intelligence and dedication.
It makes sense for Big Tech to use it because they are huge companies, and often enough their engineering talent will work in very specific projects, rarely building features using a popular framework or OSS tools.
90%+ of the companies still don't use that method because they hire people that can use popular OSS tools to build software for them.
So they look instead to judge your experience in a certain framework, language or infrastructure piece.
So overall, I believe nothing has changed.
Mathematicians will still need to learn to solve integrals and remember rules without a calculator to pass their tests.
Coders that want to join Big Tech will still often have to do leetcode challenges without AI until they find a better way to filter for intelligence and drive.
6 comments
[ 3.5 ms ] story [ 32.9 ms ] threadImagine having a toolbox, with different types of tools, and not knowing anything about them. Not knowing when it's appropriate to use one type of hammer over another.
That's the outcome of this, "Let LLMs take the wheel" attitude.
Is this pro-LLM, or anti-LLM?
I agree with the truth of the statement, but it works both ways: the world is a huge toolbox, even just code is, even just iOS is.
There's far more to learn than I can even enumerate, so much so that I can't even tell you what percentage are unknown to me, 90%, 99%, or more.
LLMs are jacks of all trades and masters of none. I'm a matter of iOS coding, but I know less than jack about, say, metalwork (where I did one small project in secondary school).
Sometimes "jack of all trades" is useful. But always check their working, because it's a miracle they can ever tell fact from fiction and were definitely trained on both.
And when it comes to hammers, the machine probably does know more than me, as I can basically only functionally distinguish them by size and if they're rubber or steel, I can't say why some have wedges on the side that doesn't strike a nail while others have balls.
I use LLM tools all day but because I used to do this all by hand I know what is wrong and what isn't. A graphing calculator is never wrong, all these LLMs make a lot of mistakes.
Someone just posted on a car forum what Gemini AI told them the required torque for a wheel bolt on a specific car is supposed to be. It was off by a huge factor to what is stated in the manual which would result in the wheel falling off at some point.
It just became really important in the last decade or so with Big Tech using them to filter candidates as a proxy for intelligence and dedication.
It makes sense for Big Tech to use it because they are huge companies, and often enough their engineering talent will work in very specific projects, rarely building features using a popular framework or OSS tools.
90%+ of the companies still don't use that method because they hire people that can use popular OSS tools to build software for them.
So they look instead to judge your experience in a certain framework, language or infrastructure piece.
So overall, I believe nothing has changed.
Mathematicians will still need to learn to solve integrals and remember rules without a calculator to pass their tests.
Coders that want to join Big Tech will still often have to do leetcode challenges without AI until they find a better way to filter for intelligence and drive.