A severely under-represented fact. I feel like this is particularly true for me, especially when I'm struggling to solve a problem at work. It's hard to separate that feeling out from the rest of my life, and it frequently bleeds over into frustration or even lack of presence in my daily life. It requires conscious effort for me to "leave work at work"
I have the same problem. Part of the way I deal with it is, more or less, to promise that I will not give my best mental hours of the day to my job. Those I keep for myself. So in the mornings I read.
This has come up before, and yes, there is a lot of potential in my opinion to atrophy parts of your brain when using LLMs. Which also heavily depends on how you use them, as there are different ways to do so. It is one of the ways why I still try to use them as tools outside the process. Something I did comment on before
Where the line is drawn does differ per person. But I think it is good if people stop to think where they draw the line and why. For me it comes down to my ability to be able to judge the LLM output properly. I can only do that when I have an active and maintained knowledge and skill set to do so. The more I let AI generate code for me (or actually written text for that matter) the less I practice those skills myself. The less I practice those skills and the more I lean on the generative part the more likely it will start to go wrong somewhere down the line.
So in my mind there two very distinct ways of using LLMs. Or maybe more accurately there are two sides of a spectrum with a lot of area in between them:
The lazy approach where you let the LLM do the heavy lifting. Examples of this include:
Providing an LLM with some basic info, what you are looking for and let the LLM do most of the writing for you. It doesn't really matter if it is code or something like an essay.
Write a very rough draft, ask the LLM to clean it up and semi blindly use the output as your final result.
Using LLMs as a tool external to the process. Examples of this include:
Deciphering spaghetti code: LLMs generally are pretty good at picking apart code blocks and generally explaining the functional parts. It can be a good way to get around "wtf is this" bits of legacy code.
Code reviews of things you have written. To be frank, half the time the suggestions are trash but at the same time LLMs are also pretty good at spotting some non obvious issues.
Using an LLM as a thesaurus to ask for possible alternatives for specific words but decide for yourself which to use.
Asking an LLM to review writing but critically go over each point keeping in mind that they might miss the point. Akin to asking random people on the internet for feedback, there might be some good points in there but some of it might be trash.
In my personal, so very much anecdotal, experience the last method still requires people to be very much engaged. But, as I said before, I think a lot of people who start using LLMs sort of drift to the lazy approach very easily. Because it is very convenient and easy to slowly hand over more thinking to an LLM.
The latter I have seen happen all too often around me. An alarming increase of lazy non critical use of LLM tools by people who should know better. People who might have been a bit slower before but delivered excellent results now have started delivering trash. Code spanning dozens of line trying to solve something that should only take one line. Code that completely ignores and conventions or design paradigms put in place. Code that goes directly against security practices. Suddenly downgraded dependency versions (because the models training data doesn't include the latest version).
So to me, it very much makes sense to keep writing myself. Both text like comments here, documentation or mails as well as code. And again, that doesn't mean I will not use LLMs during that process. But I very much make an active effort to apply them as tools externally to the process and actively engage with them in ways that doesn't hand me answers over on a silver platter.
It is a continuous effort for me to make sure to use these as tools and not as generative machines. Because my brain, my skills are my most valuable asset here and regardless of what the future will bring I strongly believe I will benefit from trying to keep it sharp.
I only recently bought an llm sub for personal use. Many of my personal projects are exploring systems I’m interested in learning, so it’s not really about the end product. However, it’s very seductive when I get stuck to immediately reach for an llm.
Knowing that an llm can fix my problem and write the entire program in 30 mins is so damn tempting. That’s how it’s changing my brain and I hate it. Not only is it allowing me to work on projects I never would have been able to tackle but it’s also taking away the satisfaction of going from red to green lights.
Not only is it allowing me to work on projects I never would have been able to tackle
I'm interested in this perspective that a lot of people seem to share. Can you elaborate on what you mean by "never would have been able to tackle"? Do you mean in terms of scope of the project, or the fact that you can now do more projects in a shorter period of time?
I'm happy to chime in as someone who has this perspective: the short answer is both scope and quantity.
I make more software than I used to, and it's usually lower quality than I would have made by hand. Mostly these kinds of things are prototypes of workflows or interfaces that pop into my head and are only worth 1 or 2 hours of investigation (but might have taken me 4 or 6 to do by hand). In the past I simply wouldn't have set aside the time to investigate these ideas, but now it's pretty easy to fit into an evening and it's not a huge investment if the prototype turns out to be a dead-end.
I also find that while my idea of "large scope" isn't really all big, the help of an LLM does mean that I take on projects that are bigger than I usually would (because I know I can delegate the boilerplate and project setup and focus on other parts).
Another point about scope is that I also am more likely to work on a project that is outside my wheelhouse. For example, I don't know how to write a Firefox extension, and don't yet care to learn. Pre-LLMs, if a project I wanted to make needed an extension, I might just not do the project because I don't want to open that can of worms. With LLMs, it is pretty straightforward to fill that gap.
This all seems pretty fair, although I don't share the same instinct. I have a ton of project ideas, but for some reason I can't bring myself to use an LLM to crank out even a prototype. (Partially because I don't want to pay for it.) I think there's still something about the "manual" investigation that I still enjoy.
I also think there's a subtle thing about the word "never" from OP. It would be a shame if LLMs are causing people to think they "never" could have done a project because they were somehow incapable of it intellectually. That's a very different feeling than "never" from the perspective of time.
but now it's pretty easy to fit into an evening and it's not a huge investment if the prototype turns out to be a dead-end.
This I’ve heard a a couple of times, and I don’t fully understand the logic of it.
You’re going to live ( i presume) for thousands of evenings yet. Why abstain from projects taking more than one?
I personally set aside one three hour session every week (sundays, first few hours after waking, but I’ve also done stretches with monday evenings), and whatever I can’t do within that session, i complete the next. Or the next.
If the project «fails» then there’s always another sunday (it never truly fails, as humans we learn valuable lessons from pretty much everything).
I’m not saying you’re wrong to have that mindset. I’ve heard it enough from other people who say the same. But it’s a very different mindset from my own. I also can’t help to think about what you could have accomplished in the past, without an LLM, if you had put in the time you do now.
It definitely does something, I worked at a place that was AI-first and didn't write any code for a year. In that environment you're not only shoving a lot of at the AI, you're also expected to juggle multiple simultaneous agents working on different things to move quickly.
I've thought about what it hurt the most, and I think it's the loss of exactness. That's something we try to build up our whole careers, and it's difficult to train. A secondary loss, probably related to exactness, is understanding. For me at least, a design has to be worked through with code to really see if it will work out, how it feels to use, etc, and both inform each other. That's out the window with AI use, but there is window dressing like making the plan more detailed (but not too detailed, that's just writing code).
I haven't experienced too much "I could just have AI do it" but that's because I so over-AI'd that I'm sick of it.
nickmonad | 12 hours ago
A severely under-represented fact. I feel like this is particularly true for me, especially when I'm struggling to solve a problem at work. It's hard to separate that feeling out from the rest of my life, and it frequently bleeds over into frustration or even lack of presence in my daily life. It requires conscious effort for me to "leave work at work"
darth-cheney | 8 hours ago
I have the same problem. Part of the way I deal with it is, more or less, to promise that I will not give my best mental hours of the day to my job. Those I keep for myself. So in the mornings I read.
creesch | 10 hours ago
This has come up before, and yes, there is a lot of potential in my opinion to atrophy parts of your brain when using LLMs. Which also heavily depends on how you use them, as there are different ways to do so. It is one of the ways why I still try to use them as tools outside the process. Something I did comment on before
It is a continuous effort for me to make sure to use these as tools and not as generative machines. Because my brain, my skills are my most valuable asset here and regardless of what the future will bring I strongly believe I will benefit from trying to keep it sharp.
erock | 7 hours ago
I only recently bought an llm sub for personal use. Many of my personal projects are exploring systems I’m interested in learning, so it’s not really about the end product. However, it’s very seductive when I get stuck to immediately reach for an llm.
Knowing that an llm can fix my problem and write the entire program in 30 mins is so damn tempting. That’s how it’s changing my brain and I hate it. Not only is it allowing me to work on projects I never would have been able to tackle but it’s also taking away the satisfaction of going from red to green lights.
nickmonad | 7 hours ago
I'm interested in this perspective that a lot of people seem to share. Can you elaborate on what you mean by "never would have been able to tackle"? Do you mean in terms of scope of the project, or the fact that you can now do more projects in a shorter period of time?
henderson | 5 hours ago
I'm happy to chime in as someone who has this perspective: the short answer is both scope and quantity.
I make more software than I used to, and it's usually lower quality than I would have made by hand. Mostly these kinds of things are prototypes of workflows or interfaces that pop into my head and are only worth 1 or 2 hours of investigation (but might have taken me 4 or 6 to do by hand). In the past I simply wouldn't have set aside the time to investigate these ideas, but now it's pretty easy to fit into an evening and it's not a huge investment if the prototype turns out to be a dead-end.
I also find that while my idea of "large scope" isn't really all big, the help of an LLM does mean that I take on projects that are bigger than I usually would (because I know I can delegate the boilerplate and project setup and focus on other parts).
Another point about scope is that I also am more likely to work on a project that is outside my wheelhouse. For example, I don't know how to write a Firefox extension, and don't yet care to learn. Pre-LLMs, if a project I wanted to make needed an extension, I might just not do the project because I don't want to open that can of worms. With LLMs, it is pretty straightforward to fill that gap.
Hope that answers your question!
nickmonad | 5 hours ago
This all seems pretty fair, although I don't share the same instinct. I have a ton of project ideas, but for some reason I can't bring myself to use an LLM to crank out even a prototype. (Partially because I don't want to pay for it.) I think there's still something about the "manual" investigation that I still enjoy.
I also think there's a subtle thing about the word "never" from OP. It would be a shame if LLMs are causing people to think they "never" could have done a project because they were somehow incapable of it intellectually. That's a very different feeling than "never" from the perspective of time.
robinheghan | an hour ago
This I’ve heard a a couple of times, and I don’t fully understand the logic of it.
You’re going to live ( i presume) for thousands of evenings yet. Why abstain from projects taking more than one?
I personally set aside one three hour session every week (sundays, first few hours after waking, but I’ve also done stretches with monday evenings), and whatever I can’t do within that session, i complete the next. Or the next.
If the project «fails» then there’s always another sunday (it never truly fails, as humans we learn valuable lessons from pretty much everything).
I’m not saying you’re wrong to have that mindset. I’ve heard it enough from other people who say the same. But it’s a very different mindset from my own. I also can’t help to think about what you could have accomplished in the past, without an LLM, if you had put in the time you do now.
bontaq | 4 hours ago
It definitely does something, I worked at a place that was AI-first and didn't write any code for a year. In that environment you're not only shoving a lot of at the AI, you're also expected to juggle multiple simultaneous agents working on different things to move quickly.
I've thought about what it hurt the most, and I think it's the loss of exactness. That's something we try to build up our whole careers, and it's difficult to train. A secondary loss, probably related to exactness, is understanding. For me at least, a design has to be worked through with code to really see if it will work out, how it feels to use, etc, and both inform each other. That's out the window with AI use, but there is window dressing like making the plan more detailed (but not too detailed, that's just writing code).
I haven't experienced too much "I could just have AI do it" but that's because I so over-AI'd that I'm sick of it.
gnyeki | 3 hours ago
How did the AI first strategy work out for the product in your view? It sounds like you’re no longer there?