So, yeah. About that. A few weeks ago I was hiking with my wife. We saw another couple from out of town in the parking lot and chatted shortly and went our separate ways.
The trails here are somewhat interconnected and we ended up running back into them at a lookout. They recognised us and asked if we could ID some berries they ate.
I guess they asked ChatGPT what they were and it told then raspberries, so they ate them. They were in fact not raspberries. They were thimbleberries, which luckily are also edible, but still. They thought they tasted off, but chalked it up to just being wild vs store bought.
And yet, the fields that require thinking the most are the ones that are getting the most use out of AI. Your average factory worker is not the one that uses AI, it’s the scientist, engineers and doctors.
If your AI use is counter productive, then it’s probably either because the scenario doesn’t benefit from AI or because you’re using it wrong.
There are PLENTY of cases where AI has made a world of difference productivity wise. Since it gained the ability to reason last year, there are just so many things it can do. Right now it’s better than every human at cyber security, period. Ask anyone who got to work with mythos.
Now I’ll agree that in the end cybersecurity is not a “productive” field since nobody cares when the cat doubled in size if the mouse did the same. However just last week we’ve seen something similar happen in Math, where it solved many math problems that no human or traditional computer has ever been able to solve. To say AI is “dumb” and “useless” after reading that paper feels a bit silly.
If your AI use is counter productive, then it’s probably either because the scenario doesn’t benefit from AI or because you’re using it wrong.
It’s harder to review code than to write it. Humans are kind of of lazy, and when we see code that looks generally correct we skim. We lose focus.
So if someone uses AI to generate a lot of code, it’s hard to review. If we just trust it, we lose knowledge of how our system works. If we spend the time reviewing it, that’s harder than just doing it ourselves. This is true at any scale, but the more you use it the worse it gets.
Maybe you consider using AI for code generation “using it wrong”. If so, please tell it to management at my job who are frothing at the mouth over the idea.
If you disagree that humans don’t notice problems, did you see the mistake I added in this post? Maybe you did! But most people wouldn’t.
Even if that wasn’t a problem, de-skilling is definitely a problem long term. If the AI did produce flawless code, after a few months of using it you’d start losing the ability to code yourself. Counter productive. Unless you’re management, and don’t care about your laborers.
I agree with what you’re saying in general, but most code out there doesn’t need to be high quality. There is no reason for anyone to design a web page manually in 2026 for example, AI will do it faster and cheaper than any human can dream of. And the code itself is incapable of causing issues, it’s just markdown/css. However even for high-quality code, if you look at benchmarks the top models produce higher quality code than the majority of devs.
As for losing understanding of our code, I agree but I suppose it’s the price to pay for productivity. Besides, it’s not like I understand Joe’s 30 years old ugly code any better than if AI had written it. At least AI has the decency to include comments.
How much web development experience do you personally have? Like, from before AI tooling?
Once your website is more complicated than a static marketing site, I don’t think AI is going to do it better. Some of the problems, like slopsquatting might be around for a long time. Possibly forever.
And the code itself is incapable of causing issues, it’s just markdown/css.
…what?
However even for high-quality code, if you look at benchmarks the top models produce higher quality code than the majority of devs.
Doing the wrong thing faster is not an improvement.
At least AI has the decency to include comments.
From what I’ve seen of AI comments, it’s not really helpful. It’s illusionionary confidence.
As for losing understanding of our code, I agree but I suppose it’s the price to pay for productivity.
That’s very expensive technical debt, or a massive de-skilling of labor.
Do you own a business? If you don’t, why would you dig your own grave here?
Honestly, I’ve only done web dev around 10 years ago and found it so mind-dumbing I left after a year. I have however had AI code a dashboard last year, roughly ~20k lines of mostly front-end and have yet to have heard of any issues from the ~30 people using daily. Why then would I have done this manually in a week when the AI did it better in an hour and $5?
Why then would I have done this manually in a week when the AI did it better in an hour and $5?
Some reasons may include
once you have the skills, it won’t take a week
once you have the skills, you learn them. They’re yours, not rented
you are better able to make changes to it
you can be much more confident it doesn’t have any horrible issues like exposing keys in plain text
less odds of other issues, like redundant or strange code
Some of this comes down to “why bother learning anything?”
I don’t know what your dashboard does. I made a simple display at an old job that showed some stats about new user signups. It took like an hour or two, and I’d never used the particular front end library before.
Edit: lines of code is a poor metric* but even so 20k seems like a lot
How is it not a contradiction? You say “people who are the worst at thinking” when it’s the people who are the best at thinking that use it the most, because they get to spend more time thinking by outsourcing the low-thinking tasks to AI.
It seems you might have different assumptions on who works which jobs. Factory workers span the whole range of people who are great at thinking and people who are bad at thinking. So do scientists/engineers/doctors. I don’t think I would agree that only those “who are worst at thinking use AI the most”, but if it were the case, then evidence that scientists/engineers/doctors are the people using the tool the most does not invalidate the claim. It could just means that those who are in those jobs and are bad at it are also the ones using AI.
I find the people who are worst at thinking use AI the most.
So, yeah. About that. A few weeks ago I was hiking with my wife. We saw another couple from out of town in the parking lot and chatted shortly and went our separate ways.
The trails here are somewhat interconnected and we ended up running back into them at a lookout. They recognised us and asked if we could ID some berries they ate.
I guess they asked ChatGPT what they were and it told then raspberries, so they ate them. They were in fact not raspberries. They were thimbleberries, which luckily are also edible, but still. They thought they tasted off, but chalked it up to just being wild vs store bought.
deleted by creator
What an incredibly idiotic thing to do! Both of them thought it was okay to trust the lie machine with their lives! Darwin award candidates.
The day they do the same with mushrooms will at least amuse the medical team (or the journalists).
That’s the whole point of AI. It does the ‘thinking’ for you.
And yet, the fields that require thinking the most are the ones that are getting the most use out of AI. Your average factory worker is not the one that uses AI, it’s the scientist, engineers and doctors.
Vaguely insulting to factory workers, first.
Second, much of the AI use is counter productive. Some of it is metrics chasing.
I did not mean to insult anyone.
If your AI use is counter productive, then it’s probably either because the scenario doesn’t benefit from AI or because you’re using it wrong.
There are PLENTY of cases where AI has made a world of difference productivity wise. Since it gained the ability to reason last year, there are just so many things it can do. Right now it’s better than every human at cyber security, period. Ask anyone who got to work with mythos.
Now I’ll agree that in the end cybersecurity is not a “productive” field since nobody cares when the cat doubled in size if the mouse did the same. However just last week we’ve seen something similar happen in Math, where it solved many math problems that no human or traditional computer has ever been able to solve. To say AI is “dumb” and “useless” after reading that paper feels a bit silly.
It’s harder to review code than to write it. Humans are kind of of lazy, and when we see code that looks generally correct we skim. We lose focus.
So if someone uses AI to generate a lot of code, it’s hard to review. If we just trust it, we lose knowledge of how our system works. If we spend the time reviewing it, that’s harder than just doing it ourselves. This is true at any scale, but the more you use it the worse it gets.
Maybe you consider using AI for code generation “using it wrong”. If so, please tell it to management at my job who are frothing at the mouth over the idea.
If you disagree that humans don’t notice problems, did you see the mistake I added in this post? Maybe you did! But most people wouldn’t.
Even if that wasn’t a problem, de-skilling is definitely a problem long term. If the AI did produce flawless code, after a few months of using it you’d start losing the ability to code yourself. Counter productive. Unless you’re management, and don’t care about your laborers.
I agree with what you’re saying in general, but most code out there doesn’t need to be high quality. There is no reason for anyone to design a web page manually in 2026 for example, AI will do it faster and cheaper than any human can dream of. And the code itself is incapable of causing issues, it’s just markdown/css. However even for high-quality code, if you look at benchmarks the top models produce higher quality code than the majority of devs.
As for losing understanding of our code, I agree but I suppose it’s the price to pay for productivity. Besides, it’s not like I understand Joe’s 30 years old ugly code any better than if AI had written it. At least AI has the decency to include comments.
How much web development experience do you personally have? Like, from before AI tooling?
Once your website is more complicated than a static marketing site, I don’t think AI is going to do it better. Some of the problems, like slopsquatting might be around for a long time. Possibly forever.
…what?
Doing the wrong thing faster is not an improvement.
From what I’ve seen of AI comments, it’s not really helpful. It’s illusionionary confidence.
That’s very expensive technical debt, or a massive de-skilling of labor.
Do you own a business? If you don’t, why would you dig your own grave here?
Honestly, I’ve only done web dev around 10 years ago and found it so mind-dumbing I left after a year. I have however had AI code a dashboard last year, roughly ~20k lines of mostly front-end and have yet to have heard of any issues from the ~30 people using daily. Why then would I have done this manually in a week when the AI did it better in an hour and $5?
Some reasons may include
Some of this comes down to “why bother learning anything?”
I don’t know what your dashboard does. I made a simple display at an old job that showed some stats about new user signups. It took like an hour or two, and I’d never used the particular front end library before.
Edit: lines of code is a poor metric* but even so 20k seems like a lot
*For example, the difference between
function add(a, b) { return a + b; }and
const add = (a,b) => a + bImplying there’s some kind of contradiction here? Why would you need to out-source your “thinking” if your job doesn’t require “thinking”?
How is it not a contradiction? You say “people who are the worst at thinking” when it’s the people who are the best at thinking that use it the most, because they get to spend more time thinking by outsourcing the low-thinking tasks to AI.
It seems you might have different assumptions on who works which jobs. Factory workers span the whole range of people who are great at thinking and people who are bad at thinking. So do scientists/engineers/doctors. I don’t think I would agree that only those “who are worst at thinking use AI the most”, but if it were the case, then evidence that scientists/engineers/doctors are the people using the tool the most does not invalidate the claim. It could just means that those who are in those jobs and are bad at it are also the ones using AI.