Does handing off that sort of work to people also ruin your skills in the same way? Or are AIs fundamentally different, and if so, why? Because we have no moral or social pressure to not delegate everything?
If you are doing your job as a manager it’s the scariest career possible. There is no realistic fallback to lower paid IC work after a certain amount of years. Your job is to enable others, not do things yourself.
There are shades of grey and you try to keep skills at least honed a little here and there doing R&D and side projects - but it’s just not the same as day to day production line work.
And of course some people start at a much higher baseline of skill than others. The impact is largely the same though over time.
This is by far the single largest caution I give to skilled engineers who talk to me about moving into a management track. It’s a decision not to be taken lightly.
This is why I think even the fuss about Noam Shazeer joining OpenAI needs to be seen within a context; as good as this hire is, there is no inherent reason to believe he still brings some secret undiscovered magic that others do not have in a more current form.
I’d be curious to see alternatives ownerships structures. Like an AI-coordinated collection of guilds, unions, or co-ops. If it can’t accumulate upwards, maybe the fundamental unit of ownership will stay with the workers.
"We" depends on where you are. Countries like Japan or Germany have maintained a gray collar, rather than a blue/white collar culture for exactly that reason. You will find business owners on the workshop floors frequently because there is an understanding that divorcing management from tacit work is going destroy leadership ability. That's the basis of vocational work culture, having general expertise across all domains of your job rather than being a kind of over-specialized idiot.
it's definitely easier to catch up after some time away than it would be if you'd never developed the skills in the first place or didn't have a natural talent, but you'll definitely atrophy without exercise. every leader i've ever worked for who graduated to a purely managerial/'strategic' position and didn't keep up their IC skills eventually got pretty slow on the uptake.
i appreciate that this study was done (AI and its inverse relationship to human wellbeing is one of the biggest challenges of our time IMO) but this also seems obvious
My friend said being an engineer is not just writing code, but designing and architecting the system. I agree, but we are also moving that to AI. We are offloading the thought process to a LLM, which confirms our biases and tells us what we want to hear.
Essentially, we are also losing the software architecture and design skills because we aren't talking to other engineers/architects/designers who may approach the problen in a complete different way; we are just asking the AI to confirm our thoughts.
There’s definitely a skill to using AI but it just doesn’t generalize very well.
I still believe the slot machine analogy holds to some extent, but I can honestly say my winning percentage is at least 90% for one shot generated code now.
I think if you know it's limitations (inlcluding your own), I don't think about hoping anymore.
I should note that when I say AI, I mean the collective models from all the major providers. The most important lesson is, you need to ask around.
> There’s definitely a skill to using AI but it just doesn’t generalize very well.
This I agree with. The only way working with AI can really be benefical outside of dealing with AI is, we are visited by extremely intelligent beings that will fuck up in the weirdest ways.
> I don't think about hoping anymore.
Running the exact same prompt again that already failed doesn’t have a that high success rate, but it’s also very low effort. So IMO it’s often worth attempting.
I'd really like to know the prompt, because I did try and, this took me way too long to figure out.
> There isn’t a magic prompt I could have written before knowing the solution. A retrospective prompt can reproduce the 200 lines only because it embeds the invariants you discovered. The useful AI task would have been systematically finding counterexamples and iterating toward those invariants, not generating the final file in one shot.
For me the value with AI isn't the one shot, it is the fact that it can help you iterate faster. My biggest take awway so far with AI is it can help you fail fast. The faster you can fail, the faster you can iterate.
mine def has to be the initial impl of my python to my personal fave compiler ir. i got claude chat to write it in two sentences. 4 turns helping it rmeemver where it put shit becsuse of transcript issues
https://github.com/gitsense/gsc-cli/blob/main/internal/cli/r...
Since I don't know go, I can't say which is the best. I am also sure they would not live up to a go experts expectations. I just know that since I don't know go, I would not have been able to create the tool. I can also prove that you can easily maintain AI generated code since I have been expanding and evolving it for over 6 months with pretty much no issues.
I do think there's more than enough room to claim that LLMs are probably significantly worse than that kind of human delegation though, in part because you have such a rapid cycle time that not-incredibly-rich people can't afford from humans.
Literally: the context window.
With the human you have a window that possibly extends up to _years_. With your language model you have maybe a few megabytes which is always preceded by instructions from the model maker.
I do not believe this at all. I think you'd have to have a very limited experience working with other human beings to be able to believe this.
> and by training it better.
"Oh yea, just do it _better_. That's your problem." Perhaps some people operate without any context but most of us find the experience lacking.
You're possibly mixing it up with long-term memory which doesn't keep immediate facts and details, it's for heavily processed and compressed summaries, for the lack of a better analogy from the LLM world. You aren't keeping the entire codebase in your memory, just its highly processed and conceptualized version. This conceptualization can be somewhat emulated as an agentic loop, but it can only go so far, current models quickly lose coherency and aren't good enough to predict what's important.
Models don't need to remember more details, they need stronger processing of what they already remember.
>"Oh yea, just do it _better_. That's your problem."
I think we're talking about different things. Models can be trained better to cram more intelligence into the same amount of parameters, that's what I mean. Similarly to how your ability to learn (and perform) math depends on your prior math training.
I said "context window" not "attention window." Of course spans of attention are limited. Knowledge is not. Knowledge is often highly specific.
> You're possibly mixing it up
Not really. You've simply failed to verify your understanding of my argument and instead created something of a strawman.
> You aren't keeping the entire codebase in your memory, just its highly processed and conceptualized version
And what is your basis for this claim? Why a codebase? You don't think I can't remember an entire function? Yet actors can remember entire sets of lines for a scene? Is that just a highly processed and abstracted version in their minds? And they just run some cognitive loop to recreate dialog in real time?
> they need stronger processing of what they already remember.
What is "stronger?" More time? More memory? More compute? And how is that put to use? Why is it, when given certain prompts, that LLMs reproduce 100s of pages of directly copied and copyrighted work?
> Models can be trained better to cram more intelligence into the same amount of parameters, that's what I mean.
Cool. _How_? What is the limit of this training? How efficient is it? How many resources do you need on the input for a given increase in output? Otherwise it's just a ton of hand waving going on here.
When I'm working on a problem, I can explicitly apply knowledge I gained decades ago to that problem. Not just "oh, it's somehow encoded in the training data"; not "oh, it's all mixed in there somewhere"; I can call up the memories, understand how to apply that knowledge to the problem at hand, and do so.
Except in certain very superficial ways, and despite many misguided LLM proponents' strident insistence to the contrary, LLMs do not work much like human brains.
My read of it was "this is why humans are fundamentally different from LLMs", not "this is how you fix LLMs to be like humans".
It's still hard to see how you can argue that humans' "context window", looked at in even a vaguely similar manner to that of LLMs, is tiny.
The context window is the working memory, which is not ‘years’ for humans either.
The human context window includes actual _context_ not just _data_.
if this same effect happens for very wealthy and/or very senior executives?
Programming normally highlights this difference. LLM programming makes it much less apparent but its still there, LLM are not thinking the way humans do and therefore struggle to solve many problems humans easily solve. So letting all human programming skills rot and just use LLM will halt our progress unless we reach AGI before our programmings skills are mostly gone.
And being born wealthy requires zero skill or practice.
If that is the case, then wouldn't this whole thing be a non-issue? We lose all the skills we used to have, but we don't need them because our entire job now is interacting with AI, and that skill we will continue to develop because it is what we do all day.
I don't know if I fully agree with that. Some skills are important even if you don't need them for your day job.