The issue with this is that we don't know how it works. Generally speaking, we know how the level of abstraction that we were born with works. We might have some understanding of one or two previous levels, but that decreases the farther down you go. We might understand the next level, and some of the next after that, but eventually people will be making things that we don't have the context to understand without having to unlearn a lot of what we know now.
I'm old enough to see this process in action; I used to be young and in possession of esoteric knowledge that made me infinitely in demand and now most of the things that young people have esoteric knowledge about is things that I don't particularly care about, and I'm left with a lot of finely honed skills to solve problems that have mostly been abstracted away.
I do think there's something to be said for the big difference in the current abstraction layer jump - the loss of determinism. It is meaningfully different.
There's a big difference between "deterministic + noise" non-determinism and "intelligent agent" non-determinism. Only the former can be statistically modeled to characterize and work around the noise with any reliability. If you measure hallucination rate for one prompt it tells you nothing about hallucination rate for another prompt.
> There's a big difference between "deterministic + noise" non-determinism and "intelligent agent" non-determinism
I think I agree, but could you expand?
My point is humans have made progress on understanding systems without clean determinism. Genetics and cosmology is technically deterministic, practically speaking, but I don’t think that’s accessible to pre-modern societies. Instead, trends and tendencies were observed, acted and capitalized on.
One big difference is the former doesn't "know" it's being experimented on. Gregor Mendel was able to grow large numbers of pea plants and identify patterns in their inheritance because the non-determinism followed consistent statistical laws. A plant can't suddenly decide to follow different laws of genetics because it saw a biologist.
On the other hand, an LLM "knows" what a benchmark is, and is capable of detecting benchmark-like scenarios. A prompt that appears reliable in testing may fail unexpected in the real world. In this way, an LLM is like a stock market (also composed of intelligent agents): the act of experimenting on it changes its properties. Such systems are never reliably predictable.
In engineering you have tolerances to deal with non-determinism. ”Within these bounds” and ”given these assumptions then…” is the foundation of building something _on top of_ those things. LLMs are the same, it relies on heavily exact Turing machines as input but its output is entirely unstable. Even if you can get determinism it will never get anything resembling ”bounds” out of the box. That makes it a poor foundation for building on top of. Ie it’s not a screwdriver, it’s the monkey who’s holding it.
I do not understand the need to argue that monkeys are better than screwdrivers at screwing. Just let the monkey be the best version of a monkey.
I would argue that management is a better discipline to pull from for employing LLMs; It is better equipped to deal with non-determinism and going completely off the rails.
I think it's safe to say that computing should be mostly deterministic. I know compilers use heuristics that are stochastic, but come on. Imagine if you were given a layer of abstraction that randomly flipped bits from time to time... That's not an abstraction, it's random programming.
> Imagine if you were given a layer of abstraction that randomly flipped bits from time to time... That's not an abstraction, it's random programming.
Computing has never really been deterministic all the way down. Storage gets corrupted, RAM has soft errors, networks drop/reorder packets, schedulers race, caches go stale, distributed systems partition, query planners change plans, ...
Obviously those become tolerable because we've developed layers of contracts and understanding the bounds around them. Error correcting codes, checksums, retries, consensus, idempotency, false-positive rates, SLAs, etc.
If the abstraction is “delegate this task to a junior engineer, analyst, lawyer, designer, or support rep,” you dont expect deterministic behavior. There's review, constraints, escalation, checklists, tests, and accountability.
So now instead of improving the tools of biology so we can actually understand it deeply - we increase complexity of IT so we have to rely on muddy, side-effecty tools of biology to try to infer some of the properties of the systems we made. That's depressing.
I feel crushing sadness when I think about how humans made these wonderful deterministic machines that we can understand and manipulate to do our bidding. Within a scant few generations, though, we've managed to effectively turn them into "magic boxes" that, soon, we'll only be able to poke and prod at when they don't do what we want.
People use the word "determinism" when they really mean something akin to "linearity", i.e. the predictability of a change in input on a change in output. Compilers for example are both deterministic and "more linear" in the sense that I can tell what the output will look like given a change in input (yes yes optimizations violate this to a small degree). LLMs can be made totally deterministic, but a seemingly insignificant change in input can create a drastic change in output, which is the characterstic we don't want.
The generally used term that I'm familiar with for this is "referential transparency", that given the same inputs you'll get the same outputs every time. LLMs can be deterministic (referentially transparent) but almost none are, i.e. when they are given the exact same input they do not return the exact same output.
I would say it is unambiguously defined by the problem you're solving. For example a spellchecker is inherently probabilistic so you shouldn't need hard coded rules here like we did for years. A "bug" then would be considered "weak accuracy" not "crash" or "incorrect behavior", as it might be the case in any other layer.
They should add a term for it, maybe "embarrassingly nondeterministic"
> I would say [whether nondeterminism is bad] is unambiguously defined by the problem you're solving.
Sure. You gave an example where it can work. Another example is something like a PCB or chip layout engine.
That particular domain (layout) is NP-complete. You'll never have an exhaustive brute force search for the optimal layout, but...
You can subsequently easily check whether the produced layout meets all your acceptance criteria or not.
Another example that successfully utilizes non-determinism for good outcomes is the application of genetic algorithms to things like antenna design.
This works because (a) you have a relatively cheap fitness test; and (b) as with real evolution, the mix of combining working designs and randomly introducing mutations often eventually produces outstanding results.
Presumably, if you applied genetic algorithm techniques to, e.g., creating your LLM prompts, you could also get good results, but that would probably quickly get expensive in terms of tokens.
So we're left with people just semi-randomly modifying prompts in order to try to tweak results.
When it works, it can be amazing. When it doesn't work, it's like a brick wall.
I like your "embarrassingly nondeterministic" term, but I somewhat disagree with:
> A "bug" then would be considered "weak accuracy" not "crash" or "incorrect behavior",
When a lawyer asks an LLM for citations of cases that support his position, he is arguably doing something stupid, because embedding an assertion such as "Show me cases which support X" is just asking for hallucinatory trouble with many current LLMs.
Nonetheless, I submit that hallucinations are, by definition, "incorrect behavior" and not merely "weak accuracy."
Now, nondeterminism could certainly be useful to the lawyer, in that it could help an LLM make connections that LexisNexis might not have in their database. So asking an LLM for help with legal issues is theoretically not an insane thing to do, but the results need to be checked very carefully.
I think I made an implicit distinction there that almost always get overlooked in these discussions, between runtime inference (when you run a transformer in your product) and offline inference (where you use AI in any ways anytime before that).
Chip design, etc would go to the latter category which is actually a huge space for innovation regardless of how you do it, while I was focusing more on the former which I think a lot of discussion seems to be concerned about.
Now if you are "running" your "generated code" that's where people will have strong opinions because it conflates the two completely different processes in the worst way possible.
It's kind of a useful distinction, but the categories are really blurry.
For example, the layout software vendors provide a solution that both uses non-determinism in their final product, and also checks the results against whatever criteria you give it.
If it tells you it will work, it's probably not lying, and if it tells you it won't work, it's almost certainly not lying.
So in your own creative process of building a board, you can rely on that.
Many LLMs don't have this property at all, and honestly, it really doesn't matter whether you are using the LLM to craft a legal document or to craft code; if you don't check the output yourself, you're possibly in for a world of hurt.
> Now if you are "running" your "generated code" that's where people will have strong opinions because it conflates the two completely different processes in the worst way possible.
I think a primary driver of the attractive nuisance here is that, because the execution of the resultant code is itself a deterministic process (even if the process that produced that code it isn't) people think that they can tame the process, by some combination of automated testing and looped automated reprompting.
At some level, they may not be wrong -- computer chips themselves are built on top of stochastic atomic processes, and google famously proved in its early days that it could have reliable processes on top of shitty commodity hardware.
But one key difference is that the nondeterminism in the atomic processes or in the non-ECC memory of the white box computers is uncorrelated.
And another key difference is that, unlike the layout software vendors, so far the LLM vendors seem unable or unwilling to properly self-check their own outputs.
To me, this is kind of the canary in the coal mine. You would think that the LLM vendors have every incentive to weed out bad results, that they, more than anybody else, have the understanding of root causes and probabilities that a particular output might be bad, and have more than sufficient resources to fix this, if it's not an intractable problem.
So Occam's Razor says that (currently) this is still an intractable problem.
I agree, but having abstractions allow us to learn things in a system that are broadly applied.
This is out the door for AI generated systems. Unless extreme care is taken there is is no consistency in these new-found codebases. New paradigms are introduced left and right, because in the eye of the LLM, and the prompter, they worked.
It didn't matter that the same pattern was repeated 37 times in a slightly different manner.
So your knowledge now is no longer portable.
It used to be that you could look at code, and ask why? And usually (not always) the answer was something like: 'I tried x, y, and z, and those didn't work', either by past experience or current experimentation. (we could argue if the current experimentation should be documented).
But for LLMs they put in something complex for no other reason than that it does what is requested. Reading a string from a valid source byte by byte until you hit \0 is valid, and it'll work. But if you take a step back at, read the API docs of what you're consuming, and then consider: if the API says X, why am I testing whether that's correct?
In 50 years, Gen Z will be retired, and they're going to be complaining: "The new generations don't know how anything works! They don't know how to enter in a prompt or optimize token usage. They don't even remember what keyboards are! They just Neropath into the UltraMind and synthesize facts into their neurons without even putting them to words or reading anything. We have lost something critical!"
I'm not sure that's true - you may be over generalizing
Probably not many here know the processes needed to turn sand into silicon and the expertise needed for hand grinding fancy lenses used in lithography. But we do know those things are needed and the approximate philosophical concepts behind these needs.
I think the danger comes when or if we fully automate a lot of the low level infrastructure tasks to the point where future generations do not even have the conceptual framework for how the tech they use is created
Exactly, I find this sort of nostalgizing to be unhelpful. I am not _that_ old but I did cut my teeth on C++ and learned about manual memory management, programming language design and had a few (even at the time) esoteric courses about assembly language. I am not assembly fluent and have no desire to be - but getting an understanding of how higher level code compiles into lower level assembly/machine statements is invaluable to get a sense of how performance can be manipulated and the how impactful it can be to do "silly" things like unroll loops to minimize the number of instructions needed for some higher level operation.
Issac Newton/Bernard of Chartres said we stand on the shoulders of giants and that allows us to see further afield but it also loses the detail of the ground beneath us - modern folks will no longer have the broad expertise that someone struggling through building a computer from transistors would gain through overcoming that[1], but we should learn about it academically to garner the bits of knowledge that remain salient. It is incredibly useful when experts in these highly specific fields have the communication skills and knowledge to share what is relevant to others without overwhelming them[2] with detail.
1. Real modern consumer grade computers are far too complex to understand fully without dedicating your life (or a significant portion of it) solely to that. It is more pragmatic to allow specialization and small realm expertise instead.
2. Of course, if you want to geek out and learn the minutiae that's awesome - but we can only geek so much in our limited time here.
> Generally speaking, we know how the level of abstraction that we were born with works
I'd say it's generally true that the majority of jobs of an era deal with a similar level of abstraction, and that's why most people stay on it. However, I frame this as being born with technical debt, and it's my obligation as an engineer to understand what the previous generations have built, and where it makes sense for me to work, directionally.
I think it's more that "Some human some time has known how it works", not that "Any given human knows how it (all) works.".
But yea this glosses over a bit trial-and-error designs and, so to speak, "genetic optimization" kinds of designs where we just try random stuff and say "Hey, this works. Not sure why, but it works.".
There was a period of a century or more where steam engine design parameters were determined by experiment and gut feel. The physics of fluid dynamics required to explain why a design is optimal just did not exist yet.
There's a lot of places in history where engineering far outpaced the science required to explain it.
I disagree. For example, the Wright brothers did air foil engineering empirically without full knowledge of the underlying fluid mechanics. Early and innovative engineering will often be less based on existing scientific or mathematical knowledge, while still using some of the methods. It's still absolutely defined as engineering.
Of course it is. The science comes from reproducing what you tried and getting the same results.
You may not have a good enough model for this observation to allow you to predict the results of other starting conditions, but it’s a model nonetheless, one that narrowly applies to the conditions you tried.
It was always engineering. You're making the assumption that their theories are not rooted in anything coherent at all, but almost always this is not the case. Usually a "what if we..." thought is based on other knowledge, and there's absolutely nothing wrong with trial and error... this is what research and experimentation, areas of science, are all about.
The level of abstraction when I started was that you had to solder the components (resistors, capacitors, chips - well IC holders, chips were expensive) to the PCB board - that was "building your computer", none of this nancy-boy rubbish of "plugging stuff together" with "slots that can only take one type of card".
I'm fairly certain I know how it works. Being a physicist helps with the even-lower-level-details if you want to start talking about transistor doping, or electrical circuit theory, for example
> The issue with this is that we don't know how it works. Generally speaking, we know how the level of abstraction that we were born with works.
What? Definitely not. I went to university and my first two years were subjects where I had to understand really deep levels of abstractions. I had to build logic gates, I had to work with hardware, wires, etc. I didnt see the point back then (I never used any of that professionally). The same about algorithms, databases, and a lot of things. But now I find it valuable and thankful that my professors (and whoever designed the career) considered important topics that I had to lear.
You started with logic gates. How much EE did you do, or the actual physics that makes the transistor possible? Those are the previous (deeper) levels that people had to know before they got abstracted away.
When I took software focused computer engineering around 2010, we still had courses that took us all the way down to transistors and even the physics of P and N junctions and how that applies to CMOS. (And even some basic analog electronics.)
Did I end up an expert at those layers? Of course not, but I know the basics and I know enough that if I need to I know where to start learning more. Just like I wasn't a C++ or hard realtime expert after university either, but now a decade and a half later I am pretty good at those (and a bunch of other skills that ended up relevant to my line of work).
Basically, none of the layers are "magic" to me. Even if I don't know the details of it, I know the general principle and I know I could learn more if I need it.
(I think you naturally end up an expert at the layer(s) you work in, and the knowledge tapers off as you go down (or up) the stack. For example, I know a fair bit about how the CPU works (cache coherency, pipeline stalls etc), I can passably read x86 assembly, etc. Because they affect the layer I work at (hard realtime systems C++ and now also Rust). I know far less about web dev than hardware.)
Similar experience in the 90s, but we don't really know the intricacies of doping silicone, or smelting metal to make the pins. And what about mining it?
I think the last time people knew how things were made was in preindustrial societies because they had to build everything themselves (whatever little things they had)
I partially disagree. I know the basic high level concept of all those. Would I be able to reproduce state of the art results on my own? Obviously not.
But a core part of the engineer or scientist mindset is curiosity for the sake of curiosity. Just the fact that I don't know something is enough reason for me to poke at it or otherwise learn more. Same reason I still take apart broken electronics as an adult and try to find the fault (and sometimes even repair it).
By the way, mining silicon is particularly easy: it's basically sand. The difficult part is purifying it, especially to the levels needed for modern nm scale chips.
A more useful question than making high end silicon would be: could you with reasonable tooling reproduce basic electric components? I'm talking things like light bulbs, resistors, generators, perhaps capacitors even? Just the basics crappy versions, not modern highly optimised surface mount components. And I think the anwer is yes (for me personally) if I had access to metal wire and sheet stock and industrial revolution era tools.
I was once in a museum in Bolzano, Italy. And they had the ice man mummy there (we call him Ötzi). He died like 5000years ago. And his "axe" was made of copper from some mine very far away. So even this guy probably did not know how everything he had was made.
There is a theory that he was probably very rich. So maybe less rich people were more in touch with what they owned. But still I found it fascinating that even so far back people relied on technology and materials that they didn't really know about.
I have similar experience with the poster, and the way I read it is, "from the things I build here are some examples". I did learn about advanced physics topics that enable transistors, and even did some experiments, but for fundamental stuff you "don't build stuff".
Did I do all physics or all electronic circuit design or all software stacks? Definitely not. But I spent 3 years learning (and building) about lots of stuff.
> You started with logic gates. How much EE did you do, or the actual physics that makes the transistor possible?
As someone with an EE degree, in first year studies we had physics classes on quantum topics that talked about electron tunnelling and its applicability to transistors:
The one thing I appreciated about my Computer Engineering undergrad - and it took me a few years to fully appreciate it - is that yes, we did cover those levels.
The first two years were shared with Electrical Engineering. The second two years started to specialize towards Computer Engineering topics.
* Physics and chemistry.
* Circuits.
* Transistors.
* Logic gates.
* FPGAs.
* Assembly.
* Compilers.
* CPU and hardware design.
* Operating systems.
* Networking layers.
* Programming languages.
* Computer graphics.
Did I master all of the above - absolutely not. I loved some of them, struggled with others. Generally the cut-off for how my brain works is logic gates, I was never strong at the levels below that.
But we did cover them, and I could honestly say I had at least a rough understanding and mental map of everything that happens inside a computer from the point where it's plugged into an outlet, to the point where pixels show up on the screen.
how did you make the transistors? made your own vacuum chambers?
i had to make logic gates and so on, but i wouldnt say i really learnt it, even if back in highschool i learned all the different things a 555 timer can do
Please get used to this sort of depressive, absurd and out of touch tone from HNers, it’s literally all they do now. Don’t bother calling people here hackers anymore, they have checked out emotionally and spiritually.
What's absurd is thinking because you took a logic class and made a flip flop 30 years ago that that is the ground floor and that it means you "understand how it all works". You're not building a CPU from logic gates and you don't know how it works. If I put you or OP in a room for a year I highly doubt you could build an 8 bit Atari-like CPU from scratch. I worked with wires and logic too but I'm not arrogant enough to say that I know how it all works.
Don't you think there is a difference from "knowing how it works" and "reproducing every aspect of it at the level of the state of the art" ?
Also, your example seems flawed if you restrict to a certain product. Can I build a compiler from scratch? Yes. Can I reproduce in a year a compiler with LLVM/GCC performance level? No. Can I build a compiler from scratch in a year from a room if I need to starting mining from metals, building transistors, then building the first assembler and then implementing the compiler? You can imagine the answer.
Not really, at least not in this discussion. An assembly programmer gets logic gates, a C programmer gets assembly, etc etc. And each one can say to the one above "well you don't really get it if you can't bang out some meaningful work". Otherwise idk what we're even talking about here. Having a vague understanding of something isn't "knowing how it works".
Yeah, but if we use your criteria, people have to invent their own atoms out of individual electrons, protons, neutrons, quarks, etc before you'd admit that "hey, they did build a compiler from scratch".
I think a fair bar for "from scratch" is stopping at the point they need an expensive fab process (or how else will they reproduce the workings of semiconductors?)
> If I put you or OP in a room for a year I highly doubt you could build an 8 bit Atari-like CPU from scratch.
I believe I can, depending on what "from scratch" means. With nothing but transistors, resistors, inductors, coils, capacitors, I can probably do apretty poor general-purpose CPU. Maybe something like a stack-based ISA or, if I had more time, an accumulator-based processor like a 6502, but with an 8-bit bus.
If you're going to ask me to create transistors themselves, well I believe that needs specialised equipment.
The parent post's argument can be boiled down to "You don't know absolutely everything therefore it's fine to know little or nothing." It's the Chewbacca defense of AI boosterism.
That was one the major criticism I have of my CS degree's program. Perhaps the knowledge I wish for is not truly CS in the mathematical/theoretical perspective, but I believe it would have been absolutely valuable to learn.
Yes, thats exactly what uni is for.. To learn all those previous centuries of stuff before you can either start contributing to the corpus yourself, or go get a job (Where you will pick up more immediately useful stuff)
We really dont understand how AI is working, even the earliest "genetic algorithms" could be incomprehensible, but computer systems in general, they're not really that complicated.. its like an audio mixing desk..it looks insanely complicated until you realise it's just the same few knobs repeated many times for many channels. High level languages, compilers, assembly, machine code, nand, mosfets. A single person really can understand it all.
Exactly, I'm over 50 and I remember all the complaints about script kiddies who looked at windows bat files as opposed to all the 'real programmers' who knew C and Assembly and used VIM and linux (which is still going strong)
A (sad/funny) anecdote. Someone told me earlier this week "Not using a coding agent today is like using Vim for development instead of an IDE, you're just lagging behind".
As someone who's been exclusively using Vim for my development (and can definitely integrate it with AI workflows), that's just an insanely silly opinion. But I guess it shows how the next generation thinks about these tools that they've heard of but never actually bothered to learn.
Script kiddies didn't look at the .bat files. They ran them without a clue how they worked. There was some respect for the people who wrote those scripts but none for users who couldn't write them
I expect 20 years from now there will be people reminiscing "remember claude.md? Back then we knew how it all worked. We had deep control. We counted tokens, chose the model..."
It is no small feat to put in words that we are losing something almost as quickly as we are gaining something. The undertone, despite leaning into nostalgia boils down to losing control and this uneasiness I feel growing daily. It is already shocking to a certain degree seeing very young people not being able to use a computer in the narrow sense because all they ever learned was touch interfaces and apps. Curated content, curated interfaces - everything that resembles some kind of hardship ironed out in thousand steps of iterations to appease the market which means the lowest common denominator.
But I also see that the people who can create the absolute most and the good things and the working things and the maintainable things nowadays are the people that have gained a tool, but not lost the knowledge of the medium we are using it on because we are tied to this old world so perfectly put under the spotlight in this blog post.
The financialization of everything is what's ruining everything. In the computer and internet realm, there's warm nostalgia from hours spent tinkering, building one's own PC, reformatting C drives due to malware, searching for "snippets" to add a forum or pimp a myspace page. But inevitably the money incentives come to dominate. It's all of our doing, we all dream of a better life to put it charitably.
Now everything is a means to a commercial end. Tinkering for fun and knowledge just isn't profitable. And it matters less and less what each our stance is on money and capital if the people that optimize for money and capital gobble up all the money and capital. Of all that's going on, the wealth gap is what's most troubling to me, closely followed because it's closely related is "post truth". I think post-truth is roughly caused by the fact that people are happy to believe what they want to believe toward some commercialized and/or idealogical end. You're much more likely to hate and blame your neighbor when you look around and you're the one not doing too well.
In my mind, 'tinkering for fun and knowledge' was never meant to be profitable - if you're tinkering with the express aim of making a profit i'm not sure you're really 'tinkering' so much as you're trying to create a product/service/output for someone else to pay for?
There's absolutely nothing wrong with doing that - I'm just not sure the 'ethos' of tinkering has anything to do with trying to make money and is usually reserved for describing someone playing about with something for their own enjoyment/fun with no desire to make money.
Now, of course some people did find that their tinkerings were able to make them money, but I think at its base it's a term I'd tend to say implies doing something for fun/for themselves, rather than doing it for profit?
In my experience there's still plenty of people out there tinkering just for their own personal satisfaction, but of course there's almost certain a whole load more people out there 'tinkering' to try make a profit.
Agreed, I'd say tinkering for profit has become the defacto optimization and tinkering for its own sake is less fashionable because it commands no attention. and everyone is obsessed with the modern currency: attention.
To be clear, I'm seeing this as an observed phenomenon, not that everyone made up their mind that they hate tinkering and love money. I just think it's getting really really hard to exist in the world as a normal person when the entire human collective is getting pumped commercialized hyper-media from all angles at all times now 100X'd by genAI bots. It's really exhausting and so you either opt-in to the game, monetize your now AI side-hustle to pay the rent, or opt-out and live in the woods. and get packages delivered by Amazon. heh.
hah, that made me laugh. To this day I'm so proud to have learned how to repeatedly just wipe my PC C drive when needed then reinstall XP or whatever. I'm just a web-dev that got their start by googling "how to make a website". As I leaned in there came a turning point where I wanted badly to switch from windows, php, scripts, to Ruby on Rails. But installing ruby at the time just didn't work. I worked up the courage - eventually - to partition my drive and install linux. After that, I got to experience a proper terminal and the rest is history.
The real problem is the philosophy that arose in the 1970s that "maximizing shareholder value" is the point of business. As a society, we've become obsessed with optimizing this one variable at the expense of all others and "growth" has become a religion. At it's core, the most important thing we should be optimizing for is "does this business improve the world while turning a robust and sustainable revenue stream" not "how much money can we make for investors". The latter will always optimize for incredible wealth disparity, enshittification, and mostly useless people like Elon Musk becoming stupidly wealthy because it's basically impossible for them to lose money no matter how bad their ideas are (once you're a certain level of rich, your passive income is just immense). I'm not saying shareholders/investors shouldn't be considered, but it can't be the only or the most important metric. We've completely lost the plot on the social and market purpose of a corporation.
> It is already shocking to a certain degree seeing very young people not being able to use a vehicle in the narrow sense because all they ever learned were the mechanical controls of the so-called automobile.
I think there's a fallacy where someone points out one instance of a larger trend which will, when taken to its logical progression, lead to an undesired effect; and then someone attempts to rebut the claim by pointing out that the trend has existed before and the undesired effect hasn't happened yet, so any concern is nugatory. I'd call it the grippery slope fallacy, complement to the slippery one: we haven't fallen down the slope yet, so we can't fall down it. What if an individual instance of ignorance is acceptable because people still need to have understanding in other areas, but if all understanding everywhere is eliminated then we all suffer?
the difference is that an automatic transmission doesn't make the car work worse. The modern UX landscape would rather board up a room because the door has a sharp handle than figure out how to make the handle less sharp
ETA: Or, to put it in car terms, we were all forced to take cabs (except for the people who were interested in driving, who became cab drivers) because car crashes happen or my sand eating neighbour couldn't tell which pedal was the brakes
Absolutely - you used to have to control the richness of the fuel mixture manually. You used to have to crank it to start it, manually interact with a clutch to shift gears, etc.
I appreciate the tactile joy of interacting with simple systems like those, but most times I just want to get where I'm going. Freeing my attention from those tasks allows me to pay more attention to the (inattentive) drivers around me, and try my best to not die.
Eventually a computer will handle driving for most of us, and we can lament about all the things we've lost there too. If you zoom out, most of us don't have an in-depth understanding of how an entire city works (power, garbage, sewage, maintenance, public services, politics, etc), and couldn't coordinate the various activities to keep it running if we had to. We live in towers of abstraction.
My read is that there does seem a clear difference between simple -> advanced machines vs simple -> "smart" machines. Nearly every smart machine is bullshit enshitification-in-waiting. Rent-seeking in-waiting. Smart tvs, smart cars with touch-screens. some would argue apple products. These things proclaim advancements but what they really do is black-box and dumb everything down to the lowest common denominator, then quite literally impose control over the air, and shove ads to you.
I'm all for just getting to where I need to go by using the appropriate tool, like a reliable car. But no not if it means foregoing the liberty of other options.
I think the dumbing down of things is a closely related yet adjacent issue. And I too absolutely despise the enshitification you’re describing.
I drove a BMW rental recently, and I swear it felt like I was driving around a toy. Such obvious HCI usability concerns, even my wife (non techie) was appalled and able to call out many obvious and stupid design decisions. It felt like a vibe coded car. A vision of the future, no doubt.
This time is different though (which has also been said every single time). But I'm worried this time it's true (also said every time). Doesn't help with the unease though.
The scale of it is certainly different, if nothing else
We have never before seen every single profession disrupted to this degree, not even the introduction of the personal computer introduced such a dramatic shift
No. No economy ever had essentially every single major company spending a significant fraction of its budget on hiring auto mechanics. Which is to say, for all the changes the automobile wrought, the role of the computer in industrialized society eclipses it tenfold, a hundredfold, a thousandfold. For an individual in many modern societies, being denied access to a car is already effectively crippling, and the idea of being denied access to computation threatens to be somehow even worse.
While the role of computers in industrialized society has been substantial, there's still a pretty good argument that the rise of motorized transportation, refridgeration, electricity, the telegram and antibiotics each exceed the impact of computer technology.
In 3-500 years we might have enough perspective on this to really judge it; I'm pretty sure that all we can say right now is "don't know".
They don't understand the dangers involved because the soft suspension and airtight cabin shield them from it. They don't have an intuitive sense of speed, force, momentum, traction etc. because they never rode a bicycle and fell off.
We are on average losing control, but you individually can choose whether to lose control or not.
Programming languages, UNIX, debuggers aren’t going anywhere. There is more to computing than what your boss demands and what is hyped on tech forums.
In fact I believe the indie/handmade scene will grow substantial if not boom, even if just as a hobby for most. Showing what you have made with your blood sweat and tears will elicit more praise and delight when you could have just asked a machine to do it all along.
There was a sweet spot with computer technologies for some decades where hobbyists could afford to experiment and even push the envelope in the nascent field of computing - similar to genetic radiation, many niches were formed and rapidly filled. The computing biome has evolved to the point where most entities are not operating at the low-level abstractions that were once the only means of interacting with the computing environment, instead they operate now at the highest levels of abstraction we are capable, so called "natural language".
"The difficulty was the knowledge. You came to know that machine the way you come to know anything that pushes back. The resistance was the whole medium. You only ever know the things that you can lose to."
We who grew up in this era formed a hands-on engineer's knowledge of these systems, built from experience and practice, learning these layers of abstraction as the bleeding edge developed. Many these days have entered into a world where there are easy answers abound, they just might not be right, and one has to gauge how much they care about correctness.
This is definitely true, and we definitely need something similar again. I've been using a game ("The Farmer Was Replaced") as a jumping-off point for teaching the kids Python, but the more I think about it, the more I think that they need some sort of hardware package similar to an old Apple //e that gives them just enough rope to hang themselves. It was easy, back in the day, to learn a ton (even assembler) on a system like that, and I feel like there's some value in rewinding the clock back to that point, forking the experience from there, and seeing what a new generation of kids will cook up.
Zach Barth and his former studio Zachtronics released several entertaining puzzle games based on idealized fake assembly languages. These are a lot of fun and useful for introducing registers and multiprocessing.
I keep coming back to the idea of taking that concept and just running with it, IE building up a toolchain alongside some concrete tasks to solve with it and just adding more detail until we reach a reasonably sophisticated level
I've been staring at 0x10c, carwars and battletech and there's a sense that I could build a sort of programming / engineering Zach like
> ... I think they need some sort of hardware package similar to an old Apple //e that gives them just enough rope to hang themselves.
Something like https://miniscript.org/MiniMicro/ perhaps? It's a virtual machine rather than real hardware, but apart from using disk image files instead of real floppy drives, the experience should be rather similar to what we* grew up on.
* Mine was a KayPro II running CP/M rather than an Apple //e, but you get the idea.
I'm one of the greybeards who has the 2400 BAUD modem negotiation tone sequences emblazened in my neurons.
For a while I've been meaning to set up some Wireguard connections among some of my systems. Being as busy as I am with work and family, I've relinquished that to Tailscale for now.
Sure, I could have sat down and jumped through the hoops to get everything set up and working across my various hosts, including network routes, firewall rules, key pairs, systemd units, and so forth. But the "cheap and easy" alternative was right there and worked (except when it forces re-authentication).
With LLM agents, I was able to effortlessly analyze my existing network and produce tailored scripts to do precisely what I wanted. All I had to do was review the scripts for potential security issues and what not. Looking at the script, there are 3 or 4 specific tweaks that needed to be made to my network routing rules given my network topology. I could have read a few man pages and iterated on the script by hand to eventually get there after maybe an hour or two of futzing.
The availability and effectiveness of the agents is simply too tempting for me. I'm not sure what this means about my skillset, or if that even matters any more. I am fairly confident that, so long as my brain still works well enough, I'll always be able to RTFM and figure things like this out myself. At this rate I wonder whether my kids will have the same ability. And I also wonder how much that will matter.
Regardless, I'm still helping them figure things out the "old way" without over-reliance on LLMs. One thing I'm fairly certain about is that failure to develop problem-solving skills can only put them in a worse position in life, no matter how capable AI becomes.
The thing is - everyone complains about AI stealing our attention and understanding. But you can just as easily use an LLM as a tool to gain a deeper understanding. It's just the default path for most folks is "Hey clanker, do the thing" rather than "Hello clanker, please tell me about how that thing works".
I've done at least a little of the latter, and it's amazing how underrated it is as an educational tool - especially for the solo individual.
That’s the “diets don’t work” line. I’m not denying that most people use AI for the path of least resistance, but I am saying that there’s more potential than just that, and it’s almost entirely overlooked in the broad debate.
Bad counterexample, c.f. the current obesity epidemic and the popularity of GLP-1s.
I get what you're saying, it's just that I think what you're saying is not overlooked, it's simply insignificant to the debate about the effect of these tools on society. Your argument comes off to me like someone saying "everyone is complaining about how these cigarettes cause cancer, but you can just as easily use lit cigarettes to warm your hands on a cold day. This is what everyone's missing in this debate."
I don't think we can change each others' minds with talk, we'll just have to see how it plays out in the coming decades.
The book/written word is also a technology that has preferences on how it’s used and transformations that it imposes on its users, and they’re much different than those of TV or computers.
> I could have read a few man pages and iterated on the script by hand to eventually get there after maybe an hour or two of futzing
I'm not even sure losing this method is such a bad thing. Tooling, build scripts, home networking -- any learning you gain from stumbling through man pages and trial-and-error at best results in some hard-won, single-purpose scripts, the knowledge will evaporate instantly b/c everything is so bespoke, "fix once and forget" until you have new gear and have to solve new problems years later.
Presenting a well-specified problem thoroughly to an LLM, then analyzing results and testing the output, is valid problem solving.
My friend's dad got a 1200 baud modem when they first came out for some obscene amount of money. Back then Compuserve used to charge by your connection speed. I was mowing lawns to pay $6.25 (~$20) an hour to play British Legends at 300 baud and he was paying ~$12 (~$38!!) an hour to connect at 1200. Plus phone bill fees, which weren't negligible back then, even for local calls!
“The knowledge is not in danger, in fact, it has never been safer. The AI models have read every manual that no human reads.”
I disagree. If you ask a model for a manual and it regurgitates that manual from its training data, it’s over-fitted. It will regurgitate something that looks like a training manual. Or whatever fits your query about training manuals.
You still have to push back on them sometimes when you spot an error. And you can only spot them if you already know what you’re looking for and should expect. Otherwise you have to ignore the output and just get the links which… could be outdated or made up as well. You’ll never know until you verify the results.
And this degrades with compression and time.
There’s no royal road. I agree that trying and getting frustrated and having to take the effort to understand something pays off in spades. I just think it’s still worth it and vastly under appreciated in this era of “everything fast, now.”
Modding communities are still going. Kids, afaik, are still playing around with hosting minecraft servers or whatever is en vogue/cool/meta/etc nowadays. DIY 8-bit computers are gaining popularity.
IMO the fact that something's become very mainstream doesn't necessarily mean it's been watered down for everybody. There will always be people with various levels of curiosity and enthusiasm.
I think so too. The others were already lost when smartphone replaced computers for the casual user.
Although I could imagine never having invested in software development if an agent could just have solved my problems. Same with learning an instrument. Very slow progress to what an AI can do today. And for most pop songs you cannot even claim that it copied more than the other generic pop songs.
Yes, if you cannot trim the code, there is a complexity barrier you won't march through.
> When I was young I fixed my parents’ computer and now that I’m older I fix computers for my kids. Are we the only generation that knows how computers work?
No, that person is just the only one in his family who knows how computers work. There's plenty of people in generations both older and younger that know too, just apparently not his parents or his kids. Some of the most impressive assembly written today is by people younger than 20. It'll be fine.
It's broadly true though, that younger people struggle with the fundamentals of computing, even if there exists specific examples of young people accomplishing exceptional feats.
We raised a generation of people on consumption-based computing devices, which means they don't develop the skills necessary to produce things using computers. Of course, we can teach these skills, but to do that, we must first acknowledge it's a skill which needs to be taught.
I agree; the situation "young people know all about computers" was maybe a twenty-five year phenomenon; before that, there weren't enough computers for it to be generally true, and at some point you no longer needed to know all about computers to successfully use a computer for the sorts of things the general population does.
"Young people know all about computers" was never true either. It was maybe true for people like you and me, the type of people who now frequent places like HN. But for the vast majority of people, computers were never all that interesting.
>"Young people know all about computers" was never true either.
That's true, but only because not all young people owned or used computers. I was born in the mid 70s and got my first computer as a young child. Back then you had to "know all about computers" to use the computer and get it to function. These days, every kid grows up surrounded by computers, with a computer in their pocket, and very few have even an inkling of how they work.
Thus "young people who use computers know all about computers" used to be a true statement, but no longer is.
It wouldn't have been true even if all young people in our generation would have had all the access to computers they would've liked. Most people just don't like tinkering with technology and that's fine!
The whole argument is pretty dumb anyway. Just try replacing "computers" with any other technology and you see how weird it starts sounding.
- "People use roads every day but almost nobody even knows how to make asphalt from scratch these days."
- "People use electricity every day but almost nobody even knows the basic equations for electric generators."
- "Children these days use refrigerators only as consumers, they don't even understand the differences between HCFC and A3 type refrigerants..."
Computer programmers despairing about how few children understand computers has strong vibes from https://xkcd.com/2501/ IMO.
When I look around at my generation it's all too noticeable that most of them are barely capable of using a computer for more than watching youtube either. And both generations older and younger than mine have generated exceptional programmers.
Whatever dictates if people are capable of creative work with computers, year of birth does not seem to be it.
> It was a Minecraft convention. We had the game set up with a keyboard, and a controller. By the 2nd day we realized, none of the kids could use a keyboard. So on the 2nd day we set up two controllers instead.
> Then I noticed something else. We counted it. 50% of the kids would come up to the console, push the controller out of the way, and try to touch the screen.
When I was growing up, my school had Macintosh Computers, and their displays were Soft and Squishy, and we'd spend long periods of time just squishing the displays to observe the satisfying visual effects that this would produce.
maybe another angle to this that I observed: my working class dad who had little education happily taught himself a bit of programming with computer magazines in the 80s an 90s. He actually could figure out his way around a command prompt.
I went on to study CS and teach undergrad courses and what I noticed, this started already maybe a decade ago, is that CS students who we handed bootable linux usbs couldn't figure out how to set their system up. And not just that, they just kept emailing us with statements like "it didn't work, what do I do?". It's not just lack of knowledge but complete helplessness when something doesn't work in 2 minutes. That's the biggest problem with this reliance on ChatGPT or whatever.
I think the young generation is in an even worse position. Not only do they not know how computers work, they don't even have the basic DIY problem solving attitude our parents have.
I resonate with every example given - setting jumpers by hand, sound card interrupts, autoexec.bat. I'm also a happy user of LLMs and agents. This article captured for me what is lost - which, as others point out, has long since been lost, if ever had, in other fields (e.g. modern cars vs. the Model T). I wouldn't go back, but I can still have a sense of loss.
> You opened files like autoexec.bat and you read them. Sometimes you built a boot disk for a single game. A floppy whose entire reason to exist was to start the computer in the configuration that one program demanded.
Is the premise here correct? I'm not sure that I'm convinced that a 1990s computer user who knew how to edit autoexec.bat or insert a floppy to boot their computer "knew how it worked" in a meaningful sense.
The stack of abstractions is deeper now, and all indications suggest it's not going to stop deepening. But I think the abstractions were already quite deep by the 1990s.
(I think the classic error here is a demographic one: computer nerds always poke through abstractions, because they do it for fun. I don't think that's going to stop, anymore than web browsers stopped people from writing kernels. If anything, we write more low-level code than ever, because access to the prerequisite knowledge is less gatekept than before.)
The most vocal demographic on HN right now is not the nerds. The entrepreneurs and grifters are desperate to find footing somewhere. They continue to double down on the alleged impact of "AI" and are now asking that their followers remember the moments they are most nostalgic for, assume everyone else was doing what they were back then (fumbling and confused), and ignore all the rest of history.
I doubt you ever understood the solid state physics, semiconductor fabrication processes, supply chain logistics, monetary policy, shipping routes, mining engineering, etc. "Knowing how things work" is a stone-age attitude.
Obviously no one knows every detail all the way down the stack. It's not possible, but having some level of understanding of how a system operates is useful especially if you ever have to fix it yourself or explain why/how it is broken.
Every domain you've enumerated is but a niche. Computing domain is a behemoth next to other domains, since computing itself is at the center of it all when the software is eating the world. Yes, they have their importance, but they are also mere bricks that built the current state of the world, next to computing becoming the mortar that holds them together. Dismissing computer illiteracy as inconsequential, (or at least comparing its importance to something like "shipping routes") in a world where computing systems are (in one way or another) everywhere, it's... just puzzling!
Obviously it's impossible to understand the whole stack. My rule of thumb is to understand at least two layers of abstraction above and below the thing I'm actively working on.
> The graybeards are aging out, nobody compiles their kernel anymore, and someday something deep will break and there will be no one left who can climb down and fix it. Maybe. But I think competence is the part that’s fine.
> knew a beige computer in 1995 that wouldn’t run a game until I had rearranged its bits by hand. More dependent than ever
If you look at previous article from this author, it says how Mac is amazing and how Linux sucks. Kids like that in 1990ties would buy expensive consoles, and would not deal with hack PC's to get free games.
Many people today are still dealing with cheap shitty hardware, 7 years old Android phones and sketchy ROMs... Just because there is no other option!
That other post is just plain wrong. LLMs are just about the only thing that can meaningfully benefit from having access to almost all the code your environment is built from. If you're on Linux, the agents not only have all that code accessible in web search, but also almost certainly include most of it in their training data.
If anything, agents are an opportunity for the "Linux year on the desktop". Granted, I don't personally believe Linux needs to have that year; it's just fine without any mass adoption there. It's simply no longer that important a segment. I thought most of the debate around that also died out years ago (perhaps even a decade?)
Still, I recently had the opportunity to compare how Codex configures macOS and Fedora for the same task, and the latter was a bit faster and significantly cheaper (in tokens). That's obviously anecdata - maybe the particular task was overrepresented on one side or something - but I wouldn't be surprised if this was a more general trend.
>The modem sang its negotiation out loud, and that shriek was two machines arguing about how to speak to each other, and after enough times you could hear when the argument was going south and the call was about to drop. You set little jumpers on drives with your fingernail. You knew which interrupt your sound card answered to, because if you guessed wrong it answered to nothing. The machine was made of edges and you cut yourself on them, and that is how you found out where they were.
Wow. I had literally all of these experiences. I remember how you used to have to remember the Sound Blaster interrupt settings and pass them to any game that used it. And I'd completely forgotten about IDE jumpers, but yep - how fiddly it was when you wanted to put 4 hard drives in the same computer, figure out which were slaves and masters, etc etc.
I think back to this 2019 talk about software engineering by Jonathan Blow, entitled ”Preventing the Collapse of Civilization”, more and more often lately: https://www.youtube.com/watch?v=ZSRHeXYDLko
>The knowledge is not in danger, in fact, it has never been safer. The AI models have read every manual that no human reads. They will recite, flawlessly and forever, exactly how all machines work. If this were only about competence, it would be the most secure moment in the history of computing.
Obviously this person doesn't have enough experience with the current crop of AIs. This is the exact reason why we should be concerned in the first place. Someone has to design and build the next version of everything, and AI isn't qualified to do the job, and might never be.
>What is dying is acquaintance. The plain, unglamorous intimacy of having fought a particular machine, and lost, and gone back, and finally felt the thing give.
This is true. But then again, who among us had to get up in the middle of the night, and trudge out to an outhouse? Who had to shovel a bit more coal into the furnace? Who had to pump water, and heat it on the stove? Who had to split wood, and stack it for the winter?
There are many, many things we're no longer acquainted with, and that's sad, but still ok. What's not ok is losing the competence required to maintaining the infrastructure and supply chain supporting society and civilization.
I came into the US tech industry in 2011 from a CS program (programming on my own longer), which was basically a pipeline into Java or web dev at that time (the classes on other things like OS, compilers, and FPGA were a thing, but seen by most as a hurdle not a promising or practical career pathways).
Most of the people I dealt with for the decade following that (i.e. pre-LLM) had little clue how any part of computers worked, and I don't intend that to be particularly pejorative, it's just that it's been somewhat easy to carve out a niche doing something to do with programming while having little understanding of the small or big picture.. all heuristics and applying working patterns.. for a few decades.
You can still understand whatever you want, today, with added tools. It's just a choice to turn your brain on or off. I think LLMs are perfectly fine as a learning tool to interrogate a subject, do comparatives, and then formalize your understanding by reading the sources. At a macro level brain rot is real, but it cuts across all generations and it started long before LLMs.
> At a macro level brain rot is real, but it cuts across all generations and it started long before LLMs.
The thing is, if there’s a way to do something easier people will generally just do that. All the nostalgia of the olden days has as a component that there were no alternatives as the example in the article highlights.
We can try to preserve the hard way to things that gets improved, but the hard way will only ever be niche if it doesn’t become extinct entirely. There is no going back.
Sure, humans are attracted to laziness by default and at its best technology must be used to lower toil and suffering (unfortunately there are unintended consequences etc for philosophical debates).
The bigger issue with brain rot is attention spans. Social media has ruined attention spans.
This might just be a personality trait of me, but I really like technology that's transparent enough that I can fix/modify/understand it. Opaque technology gives me an ick. It's not really a complexity thing -- Linux has this property of being transparent even though it's extremely complex. The way he describes old computers.. man, I loved that about them, there was nothing _secret_ going on, you weren't worried about telemetry or weird background processes. Then again, I still drive a manual transmission internal combustion car in 2026, so maybe I don't represent the world.
Then again, maybe there's light at the end of the tunnel. Even though AI boosters kind of annoy me, I do appreciate that they're tinkering and hacking and working to understand the technology on some level. I think there's always going to be some small number of people that are like this, in the same way there are still people that look at disassembly and understand CPU caches.
1: This is why I prefer console games. I just want to have fun without fighting with the machine.
2: There are plenty of people who appreciate old techniques and methods; and keep them alive. Think of going to a museum and seeing someone demonstrate an old craft or reenact how a craftsman did their job. For example, in my town there is an old, water-powered corn mill that still runs and sells corn meal.
At this point, I sound like a brand ambassador. But I highly recommend The Machine Stops by E.M. Forster. It's a story about what happens when the machine of modern life that we all rely on suddenly stops. It's been operational for hundreds of years, and there is no one left who knows how to fix it.
The thing that freaks me out about "the end of programmers" is that we're going to introduce a knowledge gap. And I'm unaware of any knowledge gaps in anything else we do. For any technology we use, we have humans who can work together to build it from dirt.
When no one knows how to program the tools that we rely on to fly our passenger planes, drive our cars, and manage our bank accounts, what happens when the LLM gets it wrong?
I wouldn't calling it learning more and more like "cycling through SoundBlaster DMA and IRQ options until the sound work". Still, there was an intrinsic curiosity that isn't as prevalent.
>To play a computer game in in the 1990s, you first had to understand how the computer worked.
As someone who played plenty of computer games as a teen in the 90s I assure you that I did not know how it worked. I learned how it worked later when I got a CS degree at university. Following the instructions from the booklet that came with the game may have helped me to get it working, but there was minimal understanding.
This is sort of the story of the telephone system of the 1950s-1970s, or electricity in the early 1900s, or cars from 1950-1980s, or airplanes from 1910-1939.
I have no idea how an electrical transformer works (well, other than the bare theory I learned in physics courses), or how power gets from the power company to my house, or how the circuits in my home are setup. I plug something in, and it works, and occasionally I throw a breaker if something is malfunctioning. There's no resistance there (pun not intended), and there shouldn't be. People got killed trying to wire their own homes.
I used to read about phone phreaks from the 1970s that could do black magic to get free long-distance phone calls. When I grew up in the 80s, that was basically gone. You picked up the phone, got a dial-tone, and called. And now it's really gone, with everyone having an encrypted cell phone connection over 5G, and your IMEI and IMSI being phoned home to every tower you connect to.
It's the nature of technology and capitalism. As the technology matures, it gets hidden away to become increasingly invisible to the end user, so you just do what you want to do with it. And then the engineering resources get spent on new problems.
> And now it's really gone, with everyone having an encrypted cell phone connection over 5G, and your IMEI and IMSI being phoned home to every tower you connect to.
people will hack enterprise pbx systems. The stream just flowed somewhere else.
>What is dying is acquaintance. The plain, unglamorous intimacy of having fought a particular machine, and lost, and gone back, and finally felt the thing give.
I think this is a universal feeling that accompanies any technological innovation. My phrasing is that new technology unbundles the thing people want from the craft that was formerly required to get it; any craft requires someone to learn and achieve through struggle.
Speak for yourself but I certainly don't know how it works.
Also, if you don't think the difficulty is the knowledge, I'm sorry but I must disagree. There has been more than one government or corporation or large institution that has tried to destroy history and knowledge in the past.
I think it's fair to say that knowledge is also at risk.
I can't help but think this type of step function change has happened with computers before.
The change from punch cards to magnetic storage certainly made it so you didn't have to "know how it works" for every single bit.
The change from machine code to a language like Fortran brought about such an abstraction that a Fortran dev didn't "know how it works" at that same level anymore.
At this point, the layers of abstractions between using a React component and something being rendered is immense. React VDOM, the real DOM, browser render engine (which sits on abstractions like ANGLE, skia, etc), calls OS APIs, which call driver APIs, and the lowest level of anything it is still C/C++ that is compiled (abstracted) to something closer to what the hardware expects.
I won't bore you with the ChatGPT output details, but it estimated at least 35 meaningful layers of abstraction between a React component and being rendered to the screen. LLMs seem to be the latest level of abstraction to make it so we "know how it works" less than before.
It's hard to articulate why, but 35 layers of deterministic abstraction feels very different from 1 layer of nondeterministic one.
In particular, I think every developer has experienced the need to jump "down the stack" to debug or understand something (even if not all the way down). Certainly, I think any senior developer should be at least conversant in the first few levels below wherever they "live". But this seemingly ends up looking fundamentally different in the interaction mode of an LLM, because you'd just ask it to jump down the stack for you
I still see this more like there will be "devs" that only know how to use LLMs and not any layer under it.
Just like there were web devs that only knew jQuery but not any actual JS. Game devs that know only Unreal Engine or Unity but nothing about DirectX, OpenGL, or Vulkan. C/C++ devs that know nothing about LLVM IR or the actual bits their compiler generates. It can also be a poor understanding of the layers in the abstractions above you too.
I find a lot of value in being able to understand levels beneath the layer I'm mostly working in. There are things, even/especially with LLMs, that I can grasp because of my deeper understanding.
But there have been plenty in our field that have only known their level of the abstraction for quite a long time. Every time there is something introduced that lowers the barrier to entry to making things, there is a version of this "but they don't know how it really works..."
I'm still nostalgic for the times when I had to enter the bootloader using the keys on front panel of the PDP-11 to then loaded the OS from paper tape. (not to mention multi pass compilers using paper tape between the phases...)
With how seriously rise of LLMs damages schools and university education, and thus next generation of comp sci graduates, at least we can hope to secure our arcane knowledge jobs for long.
This is also a fear I have, I am answering it with building bigger things with AI: what I loose in expertise in the details, I gain in the pleasure of architecturing bigger things.
We always were the only people who ever knew how it worked. In 1990 people fellow students called me to fix their computer, they had absolutely no idea how any of this worked. No. Idea. Yes, the machine was being difficult; but their reaction wasn't to fight it, or understand it. It was to call someone to do it in their stead.
Maybe the difference is more of the professionals in the field now haven't built that same muscle, as there's a broader group of people working in tech. Whereas the folks that could fix things in the 90s mainly gravitated to computers as a profession. Just random musing though I truly don't now.
I feel that things are pretty different. Although the example of interrupts and jumper settings being common knowledge is a bit of a stretch... it's still amazing to recall that MS-DOS was regarded as an end-user / consumer OS, and that, more generally, it really was regarded as totally normal to need to invest some time in learning about the system, files, directories, typing, configuring/customizing settings and network options just to be able to do what you wanted to do.
I find the current expectations around consumer "apps" to be totally infantile in comparison, where everything is now a single-purpose "app" that does exactly one thing when you push a button, and if you want something even a tiny bit different.. you can't, and that even basic things like files and settings are no longer accessible.
Compare with what the drivers/operators of the earliest automobiles were expected to know vs. what the same category of people are expected to know today.
There's nothing new about this particular progression - we've been through it in dozens of technologies already.
I think the difference is that with those technologies, not knowing is seen as a hindrance (I imagine most people wished they knew how their engine worked, that's useful) that one can simply live with, whereas with computers nowadays I see people just not care about these skills. When I do something interesting on a computer that other people see, the response isn't "huh, that's a neat skill" it's "why?". People don't care anymore. They don't see knowledge as useful or something that is beneficial to pursue
The graybeards are aging out, nobody compiles their kernel anymore, and someday something deep will break and there will be no one left who can climb down and fix it.
I dunno. I suspect it'll be slightly different in the OSS world[1]. There will be some folks who can climb down there and fix it, but since that skill is no longer valued (as in, being able to do that isn't valued by most users now, much less in some hypothetical future). And they will be expected to do that, now, without thanks. And when they say "I'm too old for this shit...you can go pound sand" because this was one unpaid insult too many, users will shrug their shoulders and load windows or buy a mac because those companies are paying & tasking people to keep things running. And by inches oss ceases to be a thing.
[1] Yes, I know there are many oss devs who get paid for their work. Many critical contributors do not.
> They will have a tool that does everything and asks for nothing, and they will be as easy with it as you are with the light switch you never once thought about.
Oh come on! Of course I wondered how a light switch work, and I then learnt. I remember taking apart broken electronics as a kid, and that later morphed into also trying to repair things. Including the computer (both hardware and software wise). I remember ending up reinstalling the OS so many times as a kid on the computer I had access to when I broke it in various ways past what I was able to fix.
Sure, not everyone will have that drive to understand how the world around them works under the hood. But for engineers and scientists I would expect a far higher percentage to have that sort of personality.
It doesn't matter if it isn't "pushing back", just that I don't understand it is enough to catch my interest and
a reason to go poking at the thing.
The thing is, yes it will still happen, but at substantially decreased rate. Groups of people banding together to solve a problem or figure a solution will slowly stop forming. Everyone will be out there learning just for themselves.
The collective knowledge of people will likely ebb. Things will be different and individuals will have to adapt.
There is a lot of useful information on how things work out there, if you want to learn. YouTube videos took the place of popular science magazines largely, and perhaps it will be some different medium tomorrow. But the knowledge is there, and just like in the past you have to go and look for it, just in different places.
There are also still courses you can take to learn practical skills, I for example took a short blacksmithing course a few years ago. A lot of fun and way harder than it looks (at least if you want to get good at it, my goal was just to give it a try and have fun, and 5 evenings was enough for that).
Folks who keep mentioning that this is no different than any previous upward leap of abstraction in human history are missing a key point that the frontier labs are certainly not missing: this is the first time in computing that you are becoming completely dependent on a _subscription service_. I don't need to know how my CPU works because it continues to work once built. Once I outsource all cognition to a billable service, I am forever and continuously in thrall to someone else's revenue strategy.
But this too shall pass, it used to be you were dependent on a subscription to just have access to a computer itself (mainframe) and now we all have one in our pockets. I’m sure some day we will be able to run something as good as Claude at home.
"To play a computer game in in the 1990s, you first had to understand how the computer worked.
So you learned. You opened files like autoexec.bat and you read them."
Ehh I dunno about that. I rarely, if ever, had to mess with any of that junk after Windows 3... I also didn't have to deal with any IRQ issues. So seems like it was already mostly abstracted in the "1990s" lol
Yeah same. I remember playing a bunch of Sierra games on my first PC as a kid on either. Most I had to deal with was installing the drivers for my SoundBlaster card from the included disk. Most I dealt with was putting in the CD, double clicking the installer and entering the product key.
That said I did run into my fair share of other problems, and that early era of personal computing and my access to machines is the only reason I work in computing/tech today. If my childhood wasn't full of tinkering with these fascinating machines, and I only ever had an iPhone or iPad, I likely would have turned out much different.
Anecdotally, I think it depends on age and family more than anything. I grew up with hand-me-down computers, games because we couldn't afford anything new. Started with win98 and games from the mid 90s, but running them in 2002.
I do remember having to look lots of things up and figuring out why some things wouldnt work. Then getting into building our own computers (because it was cheaper) and figuring out how to get halflife mods working...
Windows 3.x did not cohabitate well with certain games that required nearly the entire 640k of conventional memory to run. These are mainly games from the old DOS days; games that relied on DPMI were largely exempt. But to play certain Sierra titles, for instance, you had to set up one CONFIG.SYS configuration that loaded HIMEM.SYS and EMM386.EXE for Windows and another that left them out to free up that memory for your games. In later versions of DOS you could set up a menu to select the appropriate configuration at boot.
And of course IRQ diddling was still necessary to configure sound, network, and game controller hardware throughout the DOS era of gaming, which lasted well into the Windows 9x era.
Also, the quote didn't really apply in the 80s. Getting a game running on a C64, Spectrum, Amiga, ST or Apple II didn't involve understanding the computer.
That autoexec crap was unique to DOS PCs. DOS was a uniquely crappy experience in a lot of different ways.
Moral panic. We all have private tutors on every subject at our disposal now. It takes some special kind of mental gymnastics to conclude from this that no one will learn anything anymore henceforth. The article is just a long-winded way of saying "Kids today have it too easy", or, equivalently, "I'm getting old".
Having all the world's knowledge at our fingertips didn't seem to make us any smarter (quite the contrary, from where I'm sitting), so I'm skeptical that having private tutors at our disposal is going to. Perhaps for the ever-curious few, but for most people those "tutors" will probably just end up as another form of entertainment.
Yeah, I'm learning more than I have ever learned before. I suppose the worry is that people won't want to learn, though. That people will prompt the AI to just do things without caring about why and how they are done. That's a legitimate worry, but it's not new.
Desire to learn is deeply ingrained in humans and hard to root out. The education system does a pretty good job at it, unfortunately, but nothing ever is 100% effective.
I may not be a "typical" engineer as much anymore, in that I don't seem to be excited by knowing the intricate details of how things work as much as I enjoy knowing the capabilities of the systems that enable the projects I work on, and working at a higher level that might be more abstract and conceptual. AI enables that very well in that it enables jumping between paradigms and enabling creativity by removing a lot of the grind that was necessary in gathering technical details. I find it less appealing now to get into the weeds of how a thing works, even though at 10 I very much did. Like the author, I dug into the details of exactly which bits needed to be activated to make a particular program work. But as I aged, I was drawn to higher levels of abstraction, I think because I realized I could do more overall by piecing together bigger blocks. Yet I know colleagues who like to be involved in a very granular level of detail. I have many friends who almost revel in their knowledge that they are the ones who know precisely why something works the way it does. I think about the story of the NASA engineer on the Apollo 11 program who understood the assembly so well that he could confidently answer on the spot the moment the radar overload "1201 alarm" error flashed that the mission was still good to go. Some folks get so deep in this perspective that it almost seems like an affront to suggest they may be wrong about one of those details. And well they should, because precise people are very much needed to understand intricate systems.
But it makes me wonder how they feel with AI being an indeterminate system - one where you can never really know what the answer will be every time. Likely still correct, and getting better every day, but also possibly different every time. Added to that, it's exponential spread across the software engineering domain. For me, even as someone who likes high-level systems, I still relate to the feeling that the detailed part of me that still likes some of the internals may be less focused on in the future. Even for me that's kind of sad. But I am also curious about my colleagues who are super detail oriented - how do you feel about AI and how it changes the focus on that detail aspect?
When I was young I was so naïve. The whole world seemed to be barreling straight towards the things I liked and was interested in. Every year there were more and more nerds, more people with computers, more people playing video games. It felt so much like I was the future, that in a couple of decades everyone would be chit-chatting about what was coming, buying and comparing devices, etc. I could not have conceived that people younger than me would have less computer knowledge than me. If at twenty you had told me that I would wake up one day and teenagers no longer knew what a file extension is I would've asked you who won World War 3.
I've been thinking that there might not be new programming languages. The amount of code in the current popular ones will explode, so that's what all LLMs will be trained on.
Good luck coming up with a new language and getting enough content out there that LLMs will be fluent with it.
If true, I think that's a shame. There's plenty of innovation still to be done.
How pessimistic you are on this subject depends on how much you believe in domain transfer for these tools. No programming language sprung out of a vacuum, and it seems reasonable to think that a LLM sufficiently trained on all the C++, Haskell, and OCaml in the world could probably do a decent job at writing Rust, even if it had never seen it before, given the rust compiler as a tool and rust documentation as input
It certainly does worry me, though. As does the increasing amount of materials and manuals that seem to be written assuming the LLMs will be the (only? primary?) audience
> How pessimistic you are on this subject depends on how much you believe in domain transfer for these tools.
I tried to force various LLMs to generate many strange languages for the past 3 years. From Raku to Smalltalk, to Common Lisp and Janet, to Prolog, D, Nim, Zig, Elixir, F#, etc.
It's a mess.
They do a mostly good job with language semantics. Agents can also use standard tooling to catch and fix syntax errors. So far so good, though if a given language provides an unusual implementation for a popular feature, the model can get confused. It's usually easy to spot.
The problem is with the ecosystems. They tend to be tiny, and for new languages, they will be tiny almost by definition. As a result, the vast majority of the generated code either reimplements things from scratch or hallucinates dependencies. Since the agents iterate and the hallucinated deps don't materialize, the model fixes that... more often than not by writing from scratch. Or leaving a `// TODO:` comment.
What I found helpful is straight-up vendoring deps as submodules. Armed with grep and sources, they are much more likely to produce working code. But that also means you still need to handle the dependencies yourself, while it'll "just work" for a Kotlin/Gradle project.
It doesn't matter for all projects to the same degree, of course, but it is a pain point that will add to the difficulty (sky-high already) of a new language getting any traction, unfortunately :(
This has always been true. I never fixed my car. I knew how it worked well enough to know, hey, that sounds like its coming from the exhaust pipe. Then I took it to the mechanic. I can do basic maintenance on my bike, but I still take it to the bike shop. I have a small collection of vintage cameras, which means tracking down the few people left who know how some particular model works, might have parts. If your Synchro-Compur shutter needs parts, forget it. For most people, most of the time, the assumption has always been that someone else knows how to do that.
> The knowledge is not in danger, in fact, it has never been safer. The AI models have read every manual that no human reads. They will recite, flawlessly and forever, exactly how all machines work.
That's wrong, and that's exactly why the loss of knowledge is such a problem. LLMs do not, and cannot, actually know a single thing. They are a statistical model, not knowledge. When they give out wrong information (and they always will, by their very nature), you need someone with actual knowledge to be able to recognize the BS and correct it. But we are losing the knowledge, and unless things change we will be no better off than the people in dystopian sci-fi stories who pray to the machine god because nobody knows how it actually works.
I can already see future where there is group of people goading AI to right direction by repeatedly changing what they ask slightly. And then memorise the times when you got right incantation for seemingly right outcome. Without actually understanding or being able to reason about process in the middle. Possibly a seemingly inconsequential misspoken word or typo can lead to better or worse outcome. Or maybe just not saying "Please" will sometimes produce wrong output. and other times you must not use it...
Yes. And since LLMs can not improve knowledge - I mean, they can generate new arrangements of information, but they have no idea whether any of it real or making sense, unless humans - explicitly or through training - tell them, the more we rely on LLM knowledge the less the quality of it would be. Right now the LLMs are mainly in auxiliary role, so most of the knowledge erosion they generate is laughed at and relatively quickly corrected. But would this hold once the role of generative AIs increases? We already essentially entered the chaos period with news content - there's so much noise that it's basically impossible to know if any news message you read is true or manipulated somehow. This is going to start happening to more fundamental knowledge too, either on purpose or just by the force of the probabilistic nature of generative AI.
I was just thinking. What is the future when LLM is used for both code running itself and then also designing the hardware it runs on? Will it be sustainable? Or will there be some sort of error cascade that destroys the whole thing?
Maybe AGI is impossible with current model as it simply can not reliably improve itself... Enough errors in any part of loop will stop the progression.
This comment might have gone hard in 2023. Now it seems out of place. LLM's do hold knowledge empirically. They sometimes give out wrong information, like humans. And you need someone with better knowledge to correct it.
> But we are losing the knowledge
No we aren't and this is spreading FUD. Things have always been like this. Its called specialisation and this is how society progresses. I don't know how the supply chain worked to get the food to my table. That's why its so cheap!
There are more important battles for humanity to fight than someone trying to figure out how to configure Linux correctly for their hardware. I won't miss the need to waste time fighting the computer going away.
I recently had a run-in with security restrictions at work that only blocked the abstraction layer at the top, and were utterly surprised to discover that if you go down one level none of their restrictions did anything.
They eventually decided to keep their head in the sand and pretend like their restriction was useful.
The issue was a complete lack of knowledge of how the system actually worked one layer down.
Does anyone have a list of all the times this sentiment has been incorrect over the decades?
There's no arguing down that we have a greater amount of information about old hardware and software easily available now compared to back then, and there are many more enthusiasts as well.
This kind of cognitive dissonance has been going on in other disciplines, like music, since their inception. Guess what? There's thousands of highly talented young people on social media covering it all.
What people are actually upset about is that there isn't new "old stuff" and they can't make money off what can now be thoroughly understood by anyone.
Don't get me wrong, AI is a total shitshow and we're entering another winter, but there's also no reason to be jealous of what is making money right now.
There have always been layers of abstraction. I've been around for a while, and when I was a kid, the two choices I remember seeing were assembly code and simple semantic languages like BASIC.
Assembly seemed like too cryptic for me to really even follow and I never really did learn it, but at the time I remember people would say that assembly was easy and basically plain English compared to machine code.
As recently as fifteen or twenty years ago, I would occasionally check in and think of how unbelievably far away we had gotten from how the computer actually works. Like, you can just write "open window" and a window opens. Amazing.
Of course, those people writing machine code didn't need to really understand what P and N were in a transistor, let alone how an integrated circuit pulls it all together. And I'm not sure how much those guys knew about silicon dioxide.
The more complex things get and the more layers of abstraction there are, the more impossible it gets to really master things all the way down to first principles.
So what? People can carve out whatever chunk of the stack they want to really understand if they want to focus their lives on it. And for everyone else who's just trying to accomplish some other goal with computers as the tool, they will naturally use the highest level of abstraction and the simplest one for them to use, which is exactly what they should do.
This and it was demonstrated in Star Trek The Next Generation: Season 1 Episode 17 "When The Bough Breaks" [1]. They did not know how to fix their computer system or even where it resided. It did everything for them but was also killing them and making them sterile. There was nobody left to figure this out until the Enterprise crew happened along. This could happen on Earth. Who will save us?
Please write your blog posts yourself. I don't want to have to parse through this stupid claude output to get to your point. Literally just posting the prompt would be better than this.
It reads like marketing speak is the problem. Why is it written in such a urgent, compelling manner with those short punchy sentences? It may not be AI, but no one writes like that naturally. Maybe the author picked it up from reading a bunch of AI writing? The repetition of using you as a subject is also a red flag.
> "they will be as easy with it as you are with the light switch you never once thought about."
This metaphor doesn't really feel coherent.
> "What is dying is acquaintance."
This is a very odd topic sentence.
> "The kids coming up will not feel this as a loss, and they are not wrong not to."
Claude loves stacking big piles of negatives and arguing both sides of everything.
> "I went looking the other night for a recording of a modem connecting. People keep them around like pressed flowers."
No they don't?? People don't keep recordings of modems connecting around like pressed flowers. I, of course, understand that this is a simile. The simile just does not make sense.
> "I played one back for no reason except to hear it."
Again, note the unnecessarily florid negative phrasing. It would be much more parsimonious to write "I played one back just to hear it"
> "I know that sound by heart. People will never know the machine that just played the sound back to me. Not the way I knew the one that used to make it."
> "The difficulty was the knowledge. You came to know that machine the way you come to know anything that pushes back. The resistance was the whole medium."
The X -was- the Y, twice in short succession, to say the same thing.
The problem here is, you might be right, it makes sense. But, make it more approachable. Tell people why you think that, elaborate. Don't assume everyone is on the same page.
My problem is that my personal AI detector running on wetware just goes ping and it's hard to articulate why. I agree with the person you're replying to on all points but also, when I think something is AI, it's very much just a little brain ping that I would have to laboriously try to work backwards from.
LLM text often just is missing the cadence of human-written text. It never slows down or speeds up; it keeps a relentless pace. Sure, a human can write like that, and I'm sure people do, but I find it enervating to read, whether it was a human or an AI that wrote it.
There's an old saying of "Never Wrestle with a Pig. You Both Get Dirty and the Pig Likes It" which I'm about to fall victim to here.
The bulk of your comment history seems to be finger-wagging these kinds of slop accusations on others. Maybe you're right with the others, but I definitely know you're not right on this one.
What you're pointing out here sounds more like you just don't like the style of writing and the attempt to add some nostalgia to it, which is perfectly okay to not like. That's not something I can do much about and I'm not forcing you to read the post.
Yet you clearly read the post, and I think we'd all get more out of hearing what you think of the underlying premise rather than you circling some adjectives.
Sometimes that's me. I wish more submissions didn't read like AI. I worry that people can't detect tiresomely LLMy output and upvote it to the front page. I'm not mad or anything, I just find the internet less and less useful, because so much that circulates pings my personal AI detector (running in my skull). Sometimes I just close the tab in frustration but sometimes I comment, because it sucks.
(I don't actually care that much if it is LLM text, I care when something reads like it is, because it means everyone writing blogs using this style sounds like everyone else. it's alienating!)
It is interesting that your recent posts have much more LLM style tics than your post on the same site from 2023. Are you deliberately imitating LLM writing quirks?
Maybe it isn't deliberate! I was asking out of curiosity. This reads as extremely LLM-y to me, so if it was purely human-written I am curious about the process.
The issue with this is that we don't know how it works. Generally speaking, we know how the level of abstraction that we were born with works. We might have some understanding of one or two previous levels, but that decreases the farther down you go. We might understand the next level, and some of the next after that, but eventually people will be making things that we don't have the context to understand without having to unlearn a lot of what we know now.
I'm old enough to see this process in action; I used to be young and in possession of esoteric knowledge that made me infinitely in demand and now most of the things that young people have esoteric knowledge about is things that I don't particularly care about, and I'm left with a lot of finely honed skills to solve problems that have mostly been abstracted away.
I do think there's something to be said for the big difference in the current abstraction layer jump - the loss of determinism. It is meaningfully different.
Plenty of physical (and biological) processes are subjectively non-deterministic outside clean rooms. Doesn’t mean our ancestors couldn’t forge and selectively breed.
There's a big difference between "deterministic + noise" non-determinism and "intelligent agent" non-determinism. Only the former can be statistically modeled to characterize and work around the noise with any reliability. If you measure hallucination rate for one prompt it tells you nothing about hallucination rate for another prompt.
> There's a big difference between "deterministic + noise" non-determinism and "intelligent agent" non-determinism
I think I agree, but could you expand?
My point is humans have made progress on understanding systems without clean determinism. Genetics and cosmology is technically deterministic, practically speaking, but I don’t think that’s accessible to pre-modern societies. Instead, trends and tendencies were observed, acted and capitalized on.
One big difference is the former doesn't "know" it's being experimented on. Gregor Mendel was able to grow large numbers of pea plants and identify patterns in their inheritance because the non-determinism followed consistent statistical laws. A plant can't suddenly decide to follow different laws of genetics because it saw a biologist.
On the other hand, an LLM "knows" what a benchmark is, and is capable of detecting benchmark-like scenarios. A prompt that appears reliable in testing may fail unexpected in the real world. In this way, an LLM is like a stock market (also composed of intelligent agents): the act of experimenting on it changes its properties. Such systems are never reliably predictable.
In engineering you have tolerances to deal with non-determinism. ”Within these bounds” and ”given these assumptions then…” is the foundation of building something _on top of_ those things. LLMs are the same, it relies on heavily exact Turing machines as input but its output is entirely unstable. Even if you can get determinism it will never get anything resembling ”bounds” out of the box. That makes it a poor foundation for building on top of. Ie it’s not a screwdriver, it’s the monkey who’s holding it.
I do not understand the need to argue that monkeys are better than screwdrivers at screwing. Just let the monkey be the best version of a monkey.
You’ve clearly not seen monkeys. /sarcasm
I would argue that management is a better discipline to pull from for employing LLMs; It is better equipped to deal with non-determinism and going completely off the rails.
I think it's safe to say that computing should be mostly deterministic. I know compilers use heuristics that are stochastic, but come on. Imagine if you were given a layer of abstraction that randomly flipped bits from time to time... That's not an abstraction, it's random programming.
> Imagine if you were given a layer of abstraction that randomly flipped bits from time to time
To be fair, this isn’t a bad description of human vision.
> Imagine if you were given a layer of abstraction that randomly flipped bits from time to time... That's not an abstraction, it's random programming.
Computing has never really been deterministic all the way down. Storage gets corrupted, RAM has soft errors, networks drop/reorder packets, schedulers race, caches go stale, distributed systems partition, query planners change plans, ...
Obviously those become tolerable because we've developed layers of contracts and understanding the bounds around them. Error correcting codes, checksums, retries, consensus, idempotency, false-positive rates, SLAs, etc.
If the abstraction is “delegate this task to a junior engineer, analyst, lawyer, designer, or support rep,” you dont expect deterministic behavior. There's review, constraints, escalation, checklists, tests, and accountability.
There's a pretty cool article about trying to understand electronics and fix a radio the way we try to understand biology. https://mappingignorance.org/2014/07/02/biologists-cant-unde...
So now instead of improving the tools of biology so we can actually understand it deeply - we increase complexity of IT so we have to rely on muddy, side-effecty tools of biology to try to infer some of the properties of the systems we made. That's depressing.
I feel crushing sadness when I think about how humans made these wonderful deterministic machines that we can understand and manipulate to do our bidding. Within a scant few generations, though, we've managed to effectively turn them into "magic boxes" that, soon, we'll only be able to poke and prod at when they don't do what we want.
Our ancestors also had an infant mortality ratio of 4/10.
Determinism is way undervalued.
People use the word "determinism" when they really mean something akin to "linearity", i.e. the predictability of a change in input on a change in output. Compilers for example are both deterministic and "more linear" in the sense that I can tell what the output will look like given a change in input (yes yes optimizations violate this to a small degree). LLMs can be made totally deterministic, but a seemingly insignificant change in input can create a drastic change in output, which is the characterstic we don't want.
The generally used term that I'm familiar with for this is "referential transparency", that given the same inputs you'll get the same outputs every time. LLMs can be deterministic (referentially transparent) but almost none are, i.e. when they are given the exact same input they do not return the exact same output.
Wait, what is the light between that definition of referential transparency and determinism?
If there is something to be said, I have to ask. What specifically is there to be said?
Non-determinism is not inherently bad. We have had useful non-deterministic tools for (in computer years) several generations.
Non-determinism combined with bugginess?
That's a terrible combination. It is impossible to gradient-descent your way into a working prompt.
> Non-determinism is not inherently bad.
I would say it is unambiguously defined by the problem you're solving. For example a spellchecker is inherently probabilistic so you shouldn't need hard coded rules here like we did for years. A "bug" then would be considered "weak accuracy" not "crash" or "incorrect behavior", as it might be the case in any other layer.
They should add a term for it, maybe "embarrassingly nondeterministic"
> I would say [whether nondeterminism is bad] is unambiguously defined by the problem you're solving.
Sure. You gave an example where it can work. Another example is something like a PCB or chip layout engine.
That particular domain (layout) is NP-complete. You'll never have an exhaustive brute force search for the optimal layout, but...
You can subsequently easily check whether the produced layout meets all your acceptance criteria or not.
Another example that successfully utilizes non-determinism for good outcomes is the application of genetic algorithms to things like antenna design.
This works because (a) you have a relatively cheap fitness test; and (b) as with real evolution, the mix of combining working designs and randomly introducing mutations often eventually produces outstanding results.
Presumably, if you applied genetic algorithm techniques to, e.g., creating your LLM prompts, you could also get good results, but that would probably quickly get expensive in terms of tokens.
So we're left with people just semi-randomly modifying prompts in order to try to tweak results.
When it works, it can be amazing. When it doesn't work, it's like a brick wall.
I like your "embarrassingly nondeterministic" term, but I somewhat disagree with:
> A "bug" then would be considered "weak accuracy" not "crash" or "incorrect behavior",
When a lawyer asks an LLM for citations of cases that support his position, he is arguably doing something stupid, because embedding an assertion such as "Show me cases which support X" is just asking for hallucinatory trouble with many current LLMs.
Nonetheless, I submit that hallucinations are, by definition, "incorrect behavior" and not merely "weak accuracy."
Now, nondeterminism could certainly be useful to the lawyer, in that it could help an LLM make connections that LexisNexis might not have in their database. So asking an LLM for help with legal issues is theoretically not an insane thing to do, but the results need to be checked very carefully.
I think I made an implicit distinction there that almost always get overlooked in these discussions, between runtime inference (when you run a transformer in your product) and offline inference (where you use AI in any ways anytime before that).
Chip design, etc would go to the latter category which is actually a huge space for innovation regardless of how you do it, while I was focusing more on the former which I think a lot of discussion seems to be concerned about.
Now if you are "running" your "generated code" that's where people will have strong opinions because it conflates the two completely different processes in the worst way possible.
It's kind of a useful distinction, but the categories are really blurry.
For example, the layout software vendors provide a solution that both uses non-determinism in their final product, and also checks the results against whatever criteria you give it.
If it tells you it will work, it's probably not lying, and if it tells you it won't work, it's almost certainly not lying.
So in your own creative process of building a board, you can rely on that.
Many LLMs don't have this property at all, and honestly, it really doesn't matter whether you are using the LLM to craft a legal document or to craft code; if you don't check the output yourself, you're possibly in for a world of hurt.
> Now if you are "running" your "generated code" that's where people will have strong opinions because it conflates the two completely different processes in the worst way possible.
I think a primary driver of the attractive nuisance here is that, because the execution of the resultant code is itself a deterministic process (even if the process that produced that code it isn't) people think that they can tame the process, by some combination of automated testing and looped automated reprompting.
At some level, they may not be wrong -- computer chips themselves are built on top of stochastic atomic processes, and google famously proved in its early days that it could have reliable processes on top of shitty commodity hardware.
But one key difference is that the nondeterminism in the atomic processes or in the non-ECC memory of the white box computers is uncorrelated.
And another key difference is that, unlike the layout software vendors, so far the LLM vendors seem unable or unwilling to properly self-check their own outputs.
To me, this is kind of the canary in the coal mine. You would think that the LLM vendors have every incentive to weed out bad results, that they, more than anybody else, have the understanding of root causes and probabilities that a particular output might be bad, and have more than sufficient resources to fix this, if it's not an intractable problem.
So Occam's Razor says that (currently) this is still an intractable problem.
More than that: it's not an abstraction layer jump, because AI is not an abstraction.
I agree, but having abstractions allow us to learn things in a system that are broadly applied.
This is out the door for AI generated systems. Unless extreme care is taken there is is no consistency in these new-found codebases. New paradigms are introduced left and right, because in the eye of the LLM, and the prompter, they worked.
It didn't matter that the same pattern was repeated 37 times in a slightly different manner.
So your knowledge now is no longer portable.
It used to be that you could look at code, and ask why? And usually (not always) the answer was something like: 'I tried x, y, and z, and those didn't work', either by past experience or current experimentation. (we could argue if the current experimentation should be documented).
But for LLMs they put in something complex for no other reason than that it does what is requested. Reading a string from a valid source byte by byte until you hit \0 is valid, and it'll work. But if you take a step back at, read the API docs of what you're consuming, and then consider: if the API says X, why am I testing whether that's correct?
In 50 years, Gen Z will be retired, and they're going to be complaining: "The new generations don't know how anything works! They don't know how to enter in a prompt or optimize token usage. They don't even remember what keyboards are! They just Neropath into the UltraMind and synthesize facts into their neurons without even putting them to words or reading anything. We have lost something critical!"
True
I'm not sure that's true - you may be over generalizing
Probably not many here know the processes needed to turn sand into silicon and the expertise needed for hand grinding fancy lenses used in lithography. But we do know those things are needed and the approximate philosophical concepts behind these needs.
I think the danger comes when or if we fully automate a lot of the low level infrastructure tasks to the point where future generations do not even have the conceptual framework for how the tech they use is created
Exactly, I find this sort of nostalgizing to be unhelpful. I am not _that_ old but I did cut my teeth on C++ and learned about manual memory management, programming language design and had a few (even at the time) esoteric courses about assembly language. I am not assembly fluent and have no desire to be - but getting an understanding of how higher level code compiles into lower level assembly/machine statements is invaluable to get a sense of how performance can be manipulated and the how impactful it can be to do "silly" things like unroll loops to minimize the number of instructions needed for some higher level operation.
Issac Newton/Bernard of Chartres said we stand on the shoulders of giants and that allows us to see further afield but it also loses the detail of the ground beneath us - modern folks will no longer have the broad expertise that someone struggling through building a computer from transistors would gain through overcoming that[1], but we should learn about it academically to garner the bits of knowledge that remain salient. It is incredibly useful when experts in these highly specific fields have the communication skills and knowledge to share what is relevant to others without overwhelming them[2] with detail.
1. Real modern consumer grade computers are far too complex to understand fully without dedicating your life (or a significant portion of it) solely to that. It is more pragmatic to allow specialization and small realm expertise instead.
2. Of course, if you want to geek out and learn the minutiae that's awesome - but we can only geek so much in our limited time here.
> Generally speaking, we know how the level of abstraction that we were born with works
I'd say it's generally true that the majority of jobs of an era deal with a similar level of abstraction, and that's why most people stay on it. However, I frame this as being born with technical debt, and it's my obligation as an engineer to understand what the previous generations have built, and where it makes sense for me to work, directionally.
I think it's more that "Some human some time has known how it works", not that "Any given human knows how it (all) works.".
But yea this glosses over a bit trial-and-error designs and, so to speak, "genetic optimization" kinds of designs where we just try random stuff and say "Hey, this works. Not sure why, but it works.".
There was a period of a century or more where steam engine design parameters were determined by experiment and gut feel. The physics of fluid dynamics required to explain why a design is optimal just did not exist yet.
There's a lot of places in history where engineering far outpaced the science required to explain it.
If your design work is not based upon science and mathematics, it is not engineering.
Not saying engineering is required to build or improve a product; it is not.
"try this" is not engineering.
I disagree. For example, the Wright brothers did air foil engineering empirically without full knowledge of the underlying fluid mechanics. Early and innovative engineering will often be less based on existing scientific or mathematical knowledge, while still using some of the methods. It's still absolutely defined as engineering.
Of course it is. The science comes from reproducing what you tried and getting the same results.
You may not have a good enough model for this observation to allow you to predict the results of other starting conditions, but it’s a model nonetheless, one that narrowly applies to the conditions you tried.
> "try this" is not engineering.
It was always engineering. You're making the assumption that their theories are not rooted in anything coherent at all, but almost always this is not the case. Usually a "what if we..." thought is based on other knowledge, and there's absolutely nothing wrong with trial and error... this is what research and experimentation, areas of science, are all about.
The level of abstraction when I started was that you had to solder the components (resistors, capacitors, chips - well IC holders, chips were expensive) to the PCB board - that was "building your computer", none of this nancy-boy rubbish of "plugging stuff together" with "slots that can only take one type of card".
I'm fairly certain I know how it works. Being a physicist helps with the even-lower-level-details if you want to start talking about transistor doping, or electrical circuit theory, for example
> The issue with this is that we don't know how it works. Generally speaking, we know how the level of abstraction that we were born with works.
What? Definitely not. I went to university and my first two years were subjects where I had to understand really deep levels of abstractions. I had to build logic gates, I had to work with hardware, wires, etc. I didnt see the point back then (I never used any of that professionally). The same about algorithms, databases, and a lot of things. But now I find it valuable and thankful that my professors (and whoever designed the career) considered important topics that I had to lear.
You started with logic gates. How much EE did you do, or the actual physics that makes the transistor possible? Those are the previous (deeper) levels that people had to know before they got abstracted away.
When I took software focused computer engineering around 2010, we still had courses that took us all the way down to transistors and even the physics of P and N junctions and how that applies to CMOS. (And even some basic analog electronics.)
Did I end up an expert at those layers? Of course not, but I know the basics and I know enough that if I need to I know where to start learning more. Just like I wasn't a C++ or hard realtime expert after university either, but now a decade and a half later I am pretty good at those (and a bunch of other skills that ended up relevant to my line of work).
Basically, none of the layers are "magic" to me. Even if I don't know the details of it, I know the general principle and I know I could learn more if I need it.
(I think you naturally end up an expert at the layer(s) you work in, and the knowledge tapers off as you go down (or up) the stack. For example, I know a fair bit about how the CPU works (cache coherency, pipeline stalls etc), I can passably read x86 assembly, etc. Because they affect the layer I work at (hard realtime systems C++ and now also Rust). I know far less about web dev than hardware.)
Similar experience in the 90s, but we don't really know the intricacies of doping silicone, or smelting metal to make the pins. And what about mining it?
I think the last time people knew how things were made was in preindustrial societies because they had to build everything themselves (whatever little things they had)
I partially disagree. I know the basic high level concept of all those. Would I be able to reproduce state of the art results on my own? Obviously not.
But a core part of the engineer or scientist mindset is curiosity for the sake of curiosity. Just the fact that I don't know something is enough reason for me to poke at it or otherwise learn more. Same reason I still take apart broken electronics as an adult and try to find the fault (and sometimes even repair it).
By the way, mining silicon is particularly easy: it's basically sand. The difficult part is purifying it, especially to the levels needed for modern nm scale chips.
A more useful question than making high end silicon would be: could you with reasonable tooling reproduce basic electric components? I'm talking things like light bulbs, resistors, generators, perhaps capacitors even? Just the basics crappy versions, not modern highly optimised surface mount components. And I think the anwer is yes (for me personally) if I had access to metal wire and sheet stock and industrial revolution era tools.
I was once in a museum in Bolzano, Italy. And they had the ice man mummy there (we call him Ötzi). He died like 5000years ago. And his "axe" was made of copper from some mine very far away. So even this guy probably did not know how everything he had was made. There is a theory that he was probably very rich. So maybe less rich people were more in touch with what they owned. But still I found it fascinating that even so far back people relied on technology and materials that they didn't really know about.
Simply by the fact that you say computer engineering, you already went deeper than 99% of "computer people" in 2010
I have similar experience with the poster, and the way I read it is, "from the things I build here are some examples". I did learn about advanced physics topics that enable transistors, and even did some experiments, but for fundamental stuff you "don't build stuff".
Did I do all physics or all electronic circuit design or all software stacks? Definitely not. But I spent 3 years learning (and building) about lots of stuff.
> You started with logic gates. How much EE did you do, or the actual physics that makes the transistor possible?
As someone with an EE degree, in first year studies we had physics classes on quantum topics that talked about electron tunnelling and its applicability to transistors:
* https://en.wikipedia.org/wiki/Solid-state_physics
The one thing I appreciated about my Computer Engineering undergrad - and it took me a few years to fully appreciate it - is that yes, we did cover those levels.
The first two years were shared with Electrical Engineering. The second two years started to specialize towards Computer Engineering topics.
* Physics and chemistry.
* Circuits.
* Transistors.
* Logic gates.
* FPGAs.
* Assembly.
* Compilers.
* CPU and hardware design.
* Operating systems.
* Networking layers.
* Programming languages.
* Computer graphics.
Did I master all of the above - absolutely not. I loved some of them, struggled with others. Generally the cut-off for how my brain works is logic gates, I was never strong at the levels below that.
But we did cover them, and I could honestly say I had at least a rough understanding and mental map of everything that happens inside a computer from the point where it's plugged into an outlet, to the point where pixels show up on the screen.
Yes, they taught that too.
how did you make the transistors? made your own vacuum chambers?
i had to make logic gates and so on, but i wouldnt say i really learnt it, even if back in highschool i learned all the different things a 555 timer can do
We had a photolithography lab when I was at school. It's not that complicated to make a few npn transistors. The difficult part is doing it well.
Exactly.
Please get used to this sort of depressive, absurd and out of touch tone from HNers, it’s literally all they do now. Don’t bother calling people here hackers anymore, they have checked out emotionally and spiritually.
What's absurd is thinking because you took a logic class and made a flip flop 30 years ago that that is the ground floor and that it means you "understand how it all works". You're not building a CPU from logic gates and you don't know how it works. If I put you or OP in a room for a year I highly doubt you could build an 8 bit Atari-like CPU from scratch. I worked with wires and logic too but I'm not arrogant enough to say that I know how it all works.
Don't you think there is a difference from "knowing how it works" and "reproducing every aspect of it at the level of the state of the art" ?
Also, your example seems flawed if you restrict to a certain product. Can I build a compiler from scratch? Yes. Can I reproduce in a year a compiler with LLVM/GCC performance level? No. Can I build a compiler from scratch in a year from a room if I need to starting mining from metals, building transistors, then building the first assembler and then implementing the compiler? You can imagine the answer.
Not really, at least not in this discussion. An assembly programmer gets logic gates, a C programmer gets assembly, etc etc. And each one can say to the one above "well you don't really get it if you can't bang out some meaningful work". Otherwise idk what we're even talking about here. Having a vague understanding of something isn't "knowing how it works".
Yeah, but if we use your criteria, people have to invent their own atoms out of individual electrons, protons, neutrons, quarks, etc before you'd admit that "hey, they did build a compiler from scratch".
I think a fair bar for "from scratch" is stopping at the point they need an expensive fab process (or how else will they reproduce the workings of semiconductors?)
Code: The Hidden Language of Computer Hardware and Software
https://www.amazon.com/dp/0137909101
Explains how to build an 8 bit calculator with wires, switches, and light bulbs and then keeps on going.
This is a very cringe take.
Me and OP aren’t super heroes, we can’t do what a team of talented individuals created even if that team existed many moons ago. That isn’t the point.
We both questioned the tone and the conclusion of the comment.
> If I put you or OP in a room for a year I highly doubt you could build an 8 bit Atari-like CPU from scratch.
I believe I can, depending on what "from scratch" means. With nothing but transistors, resistors, inductors, coils, capacitors, I can probably do apretty poor general-purpose CPU. Maybe something like a stack-based ISA or, if I had more time, an accumulator-based processor like a 6502, but with an 8-bit bus.
If you're going to ask me to create transistors themselves, well I believe that needs specialised equipment.
The parent post's argument can be boiled down to "You don't know absolutely everything therefore it's fine to know little or nothing." It's the Chewbacca defense of AI boosterism.
That was one the major criticism I have of my CS degree's program. Perhaps the knowledge I wish for is not truly CS in the mathematical/theoretical perspective, but I believe it would have been absolutely valuable to learn.
Name checks out
come on now.
so even if you're right, checking out emotionally and spiritually just means more life lived. That ain't some kind of bad thing.
life is good sometimes. hard sometimes. and it's long sometimes, so give people a break.
> checking out emotionally and spiritually just means more life lived
What exactly about checking out makes it 'more living'? I sense a false dichotomy here
easy: we all need breaks. breaks are not bad. living doesn't only mean "doing actions"
I don't think OP argued for any type of relentless work.
Yes, thats exactly what uni is for.. To learn all those previous centuries of stuff before you can either start contributing to the corpus yourself, or go get a job (Where you will pick up more immediately useful stuff)
We really dont understand how AI is working, even the earliest "genetic algorithms" could be incomprehensible, but computer systems in general, they're not really that complicated.. its like an audio mixing desk..it looks insanely complicated until you realise it's just the same few knobs repeated many times for many channels. High level languages, compilers, assembly, machine code, nand, mosfets. A single person really can understand it all.
If you wish to make an apple pie from scratch you must first invent the universe
Wait a second, its layers all the way down
Exactly, I'm over 50 and I remember all the complaints about script kiddies who looked at windows bat files as opposed to all the 'real programmers' who knew C and Assembly and used VIM and linux (which is still going strong)
(ie https://www.explainxkcd.com/wiki/index.php/378:_Real_Program...)
but now also as an AI engineer we have to learn how harnesses, sys prompts, various models, tokens LLMs etc all work so a new abstraction is born..
layers changes, nerds and ultra-specialist nerds will remain
nostaliga is always for the last layer- the one you remembered from your teens and 20s.
A (sad/funny) anecdote. Someone told me earlier this week "Not using a coding agent today is like using Vim for development instead of an IDE, you're just lagging behind".
As someone who's been exclusively using Vim for my development (and can definitely integrate it with AI workflows), that's just an insanely silly opinion. But I guess it shows how the next generation thinks about these tools that they've heard of but never actually bothered to learn.
I just ignore people that can't coherently talk about the pros and cons of Vim.
Script kiddies didn't look at the .bat files. They ran them without a clue how they worked. There was some respect for the people who wrote those scripts but none for users who couldn't write them
I expect 20 years from now there will be people reminiscing "remember claude.md? Back then we knew how it all worked. We had deep control. We counted tokens, chose the model..."
> nostaliga is always for the last layer
Now there's a sentiment worth Jenny Holzerizing.
It is no small feat to put in words that we are losing something almost as quickly as we are gaining something. The undertone, despite leaning into nostalgia boils down to losing control and this uneasiness I feel growing daily. It is already shocking to a certain degree seeing very young people not being able to use a computer in the narrow sense because all they ever learned was touch interfaces and apps. Curated content, curated interfaces - everything that resembles some kind of hardship ironed out in thousand steps of iterations to appease the market which means the lowest common denominator.
But I also see that the people who can create the absolute most and the good things and the working things and the maintainable things nowadays are the people that have gained a tool, but not lost the knowledge of the medium we are using it on because we are tied to this old world so perfectly put under the spotlight in this blog post.
The financialization of everything is what's ruining everything. In the computer and internet realm, there's warm nostalgia from hours spent tinkering, building one's own PC, reformatting C drives due to malware, searching for "snippets" to add a forum or pimp a myspace page. But inevitably the money incentives come to dominate. It's all of our doing, we all dream of a better life to put it charitably.
Now everything is a means to a commercial end. Tinkering for fun and knowledge just isn't profitable. And it matters less and less what each our stance is on money and capital if the people that optimize for money and capital gobble up all the money and capital. Of all that's going on, the wealth gap is what's most troubling to me, closely followed because it's closely related is "post truth". I think post-truth is roughly caused by the fact that people are happy to believe what they want to believe toward some commercialized and/or idealogical end. You're much more likely to hate and blame your neighbor when you look around and you're the one not doing too well.
In my mind, 'tinkering for fun and knowledge' was never meant to be profitable - if you're tinkering with the express aim of making a profit i'm not sure you're really 'tinkering' so much as you're trying to create a product/service/output for someone else to pay for?
There's absolutely nothing wrong with doing that - I'm just not sure the 'ethos' of tinkering has anything to do with trying to make money and is usually reserved for describing someone playing about with something for their own enjoyment/fun with no desire to make money.
Now, of course some people did find that their tinkerings were able to make them money, but I think at its base it's a term I'd tend to say implies doing something for fun/for themselves, rather than doing it for profit?
In my experience there's still plenty of people out there tinkering just for their own personal satisfaction, but of course there's almost certain a whole load more people out there 'tinkering' to try make a profit.
Agreed, I'd say tinkering for profit has become the defacto optimization and tinkering for its own sake is less fashionable because it commands no attention. and everyone is obsessed with the modern currency: attention.
To be clear, I'm seeing this as an observed phenomenon, not that everyone made up their mind that they hate tinkering and love money. I just think it's getting really really hard to exist in the world as a normal person when the entire human collective is getting pumped commercialized hyper-media from all angles at all times now 100X'd by genAI bots. It's really exhausting and so you either opt-in to the game, monetize your now AI side-hustle to pay the rent, or opt-out and live in the woods. and get packages delivered by Amazon. heh.
> 'tinkering for fun and knowledge' was never meant to be profitable
That's why it's "for fun and knowledge" and not "for profit".
"warm nostalgia" "reformatting C drives due to malware"
Processing error.
hah, that made me laugh. To this day I'm so proud to have learned how to repeatedly just wipe my PC C drive when needed then reinstall XP or whatever. I'm just a web-dev that got their start by googling "how to make a website". As I leaned in there came a turning point where I wanted badly to switch from windows, php, scripts, to Ruby on Rails. But installing ruby at the time just didn't work. I worked up the courage - eventually - to partition my drive and install linux. After that, I got to experience a proper terminal and the rest is history.
The real problem is the philosophy that arose in the 1970s that "maximizing shareholder value" is the point of business. As a society, we've become obsessed with optimizing this one variable at the expense of all others and "growth" has become a religion. At it's core, the most important thing we should be optimizing for is "does this business improve the world while turning a robust and sustainable revenue stream" not "how much money can we make for investors". The latter will always optimize for incredible wealth disparity, enshittification, and mostly useless people like Elon Musk becoming stupidly wealthy because it's basically impossible for them to lose money no matter how bad their ideas are (once you're a certain level of rich, your passive income is just immense). I'm not saying shareholders/investors shouldn't be considered, but it can't be the only or the most important metric. We've completely lost the plot on the social and market purpose of a corporation.
> It is already shocking to a certain degree seeing very young people not being able to use a vehicle in the narrow sense because all they ever learned were the mechanical controls of the so-called automobile.
We could do this forever.
I think there's a fallacy where someone points out one instance of a larger trend which will, when taken to its logical progression, lead to an undesired effect; and then someone attempts to rebut the claim by pointing out that the trend has existed before and the undesired effect hasn't happened yet, so any concern is nugatory. I'd call it the grippery slope fallacy, complement to the slippery one: we haven't fallen down the slope yet, so we can't fall down it. What if an individual instance of ignorance is acceptable because people still need to have understanding in other areas, but if all understanding everywhere is eliminated then we all suffer?
the difference is that an automatic transmission doesn't make the car work worse. The modern UX landscape would rather board up a room because the door has a sharp handle than figure out how to make the handle less sharp
ETA: Or, to put it in car terms, we were all forced to take cabs (except for the people who were interested in driving, who became cab drivers) because car crashes happen or my sand eating neighbour couldn't tell which pedal was the brakes
It wasn't all that long ago that automatic transmissions had significantly worse reliability and fuel efficiency than manual transmissions.
Absolutely - you used to have to control the richness of the fuel mixture manually. You used to have to crank it to start it, manually interact with a clutch to shift gears, etc.
I appreciate the tactile joy of interacting with simple systems like those, but most times I just want to get where I'm going. Freeing my attention from those tasks allows me to pay more attention to the (inattentive) drivers around me, and try my best to not die.
Eventually a computer will handle driving for most of us, and we can lament about all the things we've lost there too. If you zoom out, most of us don't have an in-depth understanding of how an entire city works (power, garbage, sewage, maintenance, public services, politics, etc), and couldn't coordinate the various activities to keep it running if we had to. We live in towers of abstraction.
My read is that there does seem a clear difference between simple -> advanced machines vs simple -> "smart" machines. Nearly every smart machine is bullshit enshitification-in-waiting. Rent-seeking in-waiting. Smart tvs, smart cars with touch-screens. some would argue apple products. These things proclaim advancements but what they really do is black-box and dumb everything down to the lowest common denominator, then quite literally impose control over the air, and shove ads to you.
I'm all for just getting to where I need to go by using the appropriate tool, like a reliable car. But no not if it means foregoing the liberty of other options.
I think the dumbing down of things is a closely related yet adjacent issue. And I too absolutely despise the enshitification you’re describing.
I drove a BMW rental recently, and I swear it felt like I was driving around a toy. Such obvious HCI usability concerns, even my wife (non techie) was appalled and able to call out many obvious and stupid design decisions. It felt like a vibe coded car. A vision of the future, no doubt.
This time is different though (which has also been said every single time). But I'm worried this time it's true (also said every time). Doesn't help with the unease though.
The scale of it is certainly different, if nothing else
We have never before seen every single profession disrupted to this degree, not even the introduction of the personal computer introduced such a dramatic shift
(in recent memory). Surely the industrial revolution was a much larger disruption.
Maybe? Either way, we're still talking about one of the largest disruptions that global society has ever had
You mean as a distraction from the point being made?
No. No economy ever had essentially every single major company spending a significant fraction of its budget on hiring auto mechanics. Which is to say, for all the changes the automobile wrought, the role of the computer in industrialized society eclipses it tenfold, a hundredfold, a thousandfold. For an individual in many modern societies, being denied access to a car is already effectively crippling, and the idea of being denied access to computation threatens to be somehow even worse.
I would say you need to read some history.
While the role of computers in industrialized society has been substantial, there's still a pretty good argument that the rise of motorized transportation, refridgeration, electricity, the telegram and antibiotics each exceed the impact of computer technology.
In 3-500 years we might have enough perspective on this to really judge it; I'm pretty sure that all we can say right now is "don't know".
A person dies every 26 seconds in a road traffic incident somewhere in the world. A big part of that is people using machines they do not understand.
In what sense do they not understand it? What could we teach them that they don't already know?
They don't understand the dangers involved because the soft suspension and airtight cabin shield them from it. They don't have an intuitive sense of speed, force, momentum, traction etc. because they never rode a bicycle and fell off.
Almost no part of that is due to people using machines they do not understand.
We are on average losing control, but you individually can choose whether to lose control or not.
Programming languages, UNIX, debuggers aren’t going anywhere. There is more to computing than what your boss demands and what is hyped on tech forums.
In fact I believe the indie/handmade scene will grow substantial if not boom, even if just as a hobby for most. Showing what you have made with your blood sweat and tears will elicit more praise and delight when you could have just asked a machine to do it all along.
There was a sweet spot with computer technologies for some decades where hobbyists could afford to experiment and even push the envelope in the nascent field of computing - similar to genetic radiation, many niches were formed and rapidly filled. The computing biome has evolved to the point where most entities are not operating at the low-level abstractions that were once the only means of interacting with the computing environment, instead they operate now at the highest levels of abstraction we are capable, so called "natural language".
"The difficulty was the knowledge. You came to know that machine the way you come to know anything that pushes back. The resistance was the whole medium. You only ever know the things that you can lose to."
We who grew up in this era formed a hands-on engineer's knowledge of these systems, built from experience and practice, learning these layers of abstraction as the bleeding edge developed. Many these days have entered into a world where there are easy answers abound, they just might not be right, and one has to gauge how much they care about correctness.
This is definitely true, and we definitely need something similar again. I've been using a game ("The Farmer Was Replaced") as a jumping-off point for teaching the kids Python, but the more I think about it, the more I think that they need some sort of hardware package similar to an old Apple //e that gives them just enough rope to hang themselves. It was easy, back in the day, to learn a ton (even assembler) on a system like that, and I feel like there's some value in rewinding the clock back to that point, forking the experience from there, and seeing what a new generation of kids will cook up.
Zach Barth and his former studio Zachtronics released several entertaining puzzle games based on idealized fake assembly languages. These are a lot of fun and useful for introducing registers and multiprocessing.
I keep coming back to the idea of taking that concept and just running with it, IE building up a toolchain alongside some concrete tasks to solve with it and just adding more detail until we reach a reasonably sophisticated level
I've been staring at 0x10c, carwars and battletech and there's a sense that I could build a sort of programming / engineering Zach like
> ... I think they need some sort of hardware package similar to an old Apple //e that gives them just enough rope to hang themselves.
Something like https://miniscript.org/MiniMicro/ perhaps? It's a virtual machine rather than real hardware, but apart from using disk image files instead of real floppy drives, the experience should be rather similar to what we* grew up on.
* Mine was a KayPro II running CP/M rather than an Apple //e, but you get the idea.
If you want to go the old computer route, find an old vic-20. It has the best manual for any computer ever made imo
I'm one of the greybeards who has the 2400 BAUD modem negotiation tone sequences emblazened in my neurons.
For a while I've been meaning to set up some Wireguard connections among some of my systems. Being as busy as I am with work and family, I've relinquished that to Tailscale for now.
Sure, I could have sat down and jumped through the hoops to get everything set up and working across my various hosts, including network routes, firewall rules, key pairs, systemd units, and so forth. But the "cheap and easy" alternative was right there and worked (except when it forces re-authentication).
With LLM agents, I was able to effortlessly analyze my existing network and produce tailored scripts to do precisely what I wanted. All I had to do was review the scripts for potential security issues and what not. Looking at the script, there are 3 or 4 specific tweaks that needed to be made to my network routing rules given my network topology. I could have read a few man pages and iterated on the script by hand to eventually get there after maybe an hour or two of futzing.
The availability and effectiveness of the agents is simply too tempting for me. I'm not sure what this means about my skillset, or if that even matters any more. I am fairly confident that, so long as my brain still works well enough, I'll always be able to RTFM and figure things like this out myself. At this rate I wonder whether my kids will have the same ability. And I also wonder how much that will matter.
Regardless, I'm still helping them figure things out the "old way" without over-reliance on LLMs. One thing I'm fairly certain about is that failure to develop problem-solving skills can only put them in a worse position in life, no matter how capable AI becomes.
The thing is - everyone complains about AI stealing our attention and understanding. But you can just as easily use an LLM as a tool to gain a deeper understanding. It's just the default path for most folks is "Hey clanker, do the thing" rather than "Hello clanker, please tell me about how that thing works".
I've done at least a little of the latter, and it's amazing how underrated it is as an educational tool - especially for the solo individual.
And you can use a TV to read books, but nobody does.
Every technology has ways it can be used, and ways it wants to be used. This one wants to be used in a way that produces outcomes we won't like.
That’s the “diets don’t work” line. I’m not denying that most people use AI for the path of least resistance, but I am saying that there’s more potential than just that, and it’s almost entirely overlooked in the broad debate.
Bad counterexample, c.f. the current obesity epidemic and the popularity of GLP-1s.
I get what you're saying, it's just that I think what you're saying is not overlooked, it's simply insignificant to the debate about the effect of these tools on society. Your argument comes off to me like someone saying "everyone is complaining about how these cigarettes cause cancer, but you can just as easily use lit cigarettes to warm your hands on a cold day. This is what everyone's missing in this debate."
I don't think we can change each others' minds with talk, we'll just have to see how it plays out in the coming decades.
I think what you are describing is covered by either an E reader or the giant collection of how-to explantions on YouTube?
Videos aren’t books, and ereaders aren’t TVs.
The book/written word is also a technology that has preferences on how it’s used and transformations that it imposes on its users, and they’re much different than those of TV or computers.
Nobody has time anymore to read a textbook.
> I could have read a few man pages and iterated on the script by hand to eventually get there after maybe an hour or two of futzing
I'm not even sure losing this method is such a bad thing. Tooling, build scripts, home networking -- any learning you gain from stumbling through man pages and trial-and-error at best results in some hard-won, single-purpose scripts, the knowledge will evaporate instantly b/c everything is so bespoke, "fix once and forget" until you have new gear and have to solve new problems years later.
Presenting a well-specified problem thoroughly to an LLM, then analyzing results and testing the output, is valid problem solving.
>I'm one of the greybeards who has the 2400 BAUD modem negotiation tone sequences emblazened in my neurons.
Making the jump from 300 baud to 2400 felt like magic!
I started on a faulty 1200 BAUD modem that would only connect at 300 BAUD. And I used to load my programs off a cassette tape.
If we keep this up, someone's bound to jump in with a brag about memory cores and punchcards.
I don't think they will be jumping anywhere TBH.
(young-un with the 28.8k starter here, so I just need some stretching to keep up with you old fellas :P)
My friend's dad got a 1200 baud modem when they first came out for some obscene amount of money. Back then Compuserve used to charge by your connection speed. I was mowing lawns to pay $6.25 (~$20) an hour to play British Legends at 300 baud and he was paying ~$12 (~$38!!) an hour to connect at 1200. Plus phone bill fees, which weren't negligible back then, even for local calls!
https://www.chicagotribune.com/1987/05/31/compuserve-sees-th...
“The knowledge is not in danger, in fact, it has never been safer. The AI models have read every manual that no human reads.”
I disagree. If you ask a model for a manual and it regurgitates that manual from its training data, it’s over-fitted. It will regurgitate something that looks like a training manual. Or whatever fits your query about training manuals.
You still have to push back on them sometimes when you spot an error. And you can only spot them if you already know what you’re looking for and should expect. Otherwise you have to ignore the output and just get the links which… could be outdated or made up as well. You’ll never know until you verify the results.
And this degrades with compression and time.
There’s no royal road. I agree that trying and getting frustrated and having to take the effort to understand something pays off in spades. I just think it’s still worth it and vastly under appreciated in this era of “everything fast, now.”
Modding communities are still going. Kids, afaik, are still playing around with hosting minecraft servers or whatever is en vogue/cool/meta/etc nowadays. DIY 8-bit computers are gaining popularity.
IMO the fact that something's become very mainstream doesn't necessarily mean it's been watered down for everybody. There will always be people with various levels of curiosity and enthusiasm.
Yeah, I don't think human curiosity can be extinguished. It can be disincentivized, yes, but not extinguished. Nerds will always be nerds.
I think so too. The others were already lost when smartphone replaced computers for the casual user.
Although I could imagine never having invested in software development if an agent could just have solved my problems. Same with learning an instrument. Very slow progress to what an AI can do today. And for most pop songs you cannot even claim that it copied more than the other generic pop songs.
Yes, if you cannot trim the code, there is a complexity barrier you won't march through.
@ryancbriggs - 18 Oct 2024
> When I was young I fixed my parents’ computer and now that I’m older I fix computers for my kids. Are we the only generation that knows how computers work?
https://x.com/ryancbriggs/status/1847391612428517844
https://xcancel.com/ryancbriggs/status/1847391612428517844
No, that person is just the only one in his family who knows how computers work. There's plenty of people in generations both older and younger that know too, just apparently not his parents or his kids. Some of the most impressive assembly written today is by people younger than 20. It'll be fine.
It's broadly true though, that younger people struggle with the fundamentals of computing, even if there exists specific examples of young people accomplishing exceptional feats.
We raised a generation of people on consumption-based computing devices, which means they don't develop the skills necessary to produce things using computers. Of course, we can teach these skills, but to do that, we must first acknowledge it's a skill which needs to be taught.
I agree; the situation "young people know all about computers" was maybe a twenty-five year phenomenon; before that, there weren't enough computers for it to be generally true, and at some point you no longer needed to know all about computers to successfully use a computer for the sorts of things the general population does.
"Young people know all about computers" was never true either. It was maybe true for people like you and me, the type of people who now frequent places like HN. But for the vast majority of people, computers were never all that interesting.
>"Young people know all about computers" was never true either.
That's true, but only because not all young people owned or used computers. I was born in the mid 70s and got my first computer as a young child. Back then you had to "know all about computers" to use the computer and get it to function. These days, every kid grows up surrounded by computers, with a computer in their pocket, and very few have even an inkling of how they work.
Thus "young people who use computers know all about computers" used to be a true statement, but no longer is.
It wouldn't have been true even if all young people in our generation would have had all the access to computers they would've liked. Most people just don't like tinkering with technology and that's fine!
The whole argument is pretty dumb anyway. Just try replacing "computers" with any other technology and you see how weird it starts sounding.
- "People use roads every day but almost nobody even knows how to make asphalt from scratch these days."
- "People use electricity every day but almost nobody even knows the basic equations for electric generators."
- "Children these days use refrigerators only as consumers, they don't even understand the differences between HCFC and A3 type refrigerants..."
Computer programmers despairing about how few children understand computers has strong vibes from https://xkcd.com/2501/ IMO.
Most people in my generation didn't know the fundamentals of computing either. That's why I get paid the big bucks to make computers work for them.
>Most people in my generation didn't know the fundamentals of computing either.
That's true of my generation as well, but computers weren't ubiquitous and used by everyone all the time for virtually everything.
When I look around at my generation it's all too noticeable that most of them are barely capable of using a computer for more than watching youtube either. And both generations older and younger than mine have generated exceptional programmers.
Whatever dictates if people are capable of creative work with computers, year of birth does not seem to be it.
https://www.youtube.com/shorts/h8ElOpITBjQ
> It was a Minecraft convention. We had the game set up with a keyboard, and a controller. By the 2nd day we realized, none of the kids could use a keyboard. So on the 2nd day we set up two controllers instead.
> Then I noticed something else. We counted it. 50% of the kids would come up to the console, push the controller out of the way, and try to touch the screen.
I never felt more violated than when a coworker 15 years younger than me started to touch my work laptop screen.
I mean, she was right, it turned out to be a touch screen, but really who does something like that?
When I was growing up, my school had Macintosh Computers, and their displays were Soft and Squishy, and we'd spend long periods of time just squishing the displays to observe the satisfying visual effects that this would produce.
maybe another angle to this that I observed: my working class dad who had little education happily taught himself a bit of programming with computer magazines in the 80s an 90s. He actually could figure out his way around a command prompt.
I went on to study CS and teach undergrad courses and what I noticed, this started already maybe a decade ago, is that CS students who we handed bootable linux usbs couldn't figure out how to set their system up. And not just that, they just kept emailing us with statements like "it didn't work, what do I do?". It's not just lack of knowledge but complete helplessness when something doesn't work in 2 minutes. That's the biggest problem with this reliance on ChatGPT or whatever.
I think the young generation is in an even worse position. Not only do they not know how computers work, they don't even have the basic DIY problem solving attitude our parents have.
I resonate with every example given - setting jumpers by hand, sound card interrupts, autoexec.bat. I'm also a happy user of LLMs and agents. This article captured for me what is lost - which, as others point out, has long since been lost, if ever had, in other fields (e.g. modern cars vs. the Model T). I wouldn't go back, but I can still have a sense of loss.
Beautiful writing.
> You opened files like autoexec.bat and you read them. Sometimes you built a boot disk for a single game. A floppy whose entire reason to exist was to start the computer in the configuration that one program demanded.
Is the premise here correct? I'm not sure that I'm convinced that a 1990s computer user who knew how to edit autoexec.bat or insert a floppy to boot their computer "knew how it worked" in a meaningful sense.
The stack of abstractions is deeper now, and all indications suggest it's not going to stop deepening. But I think the abstractions were already quite deep by the 1990s.
(I think the classic error here is a demographic one: computer nerds always poke through abstractions, because they do it for fun. I don't think that's going to stop, anymore than web browsers stopped people from writing kernels. If anything, we write more low-level code than ever, because access to the prerequisite knowledge is less gatekept than before.)
This isn't said enough.
The most vocal demographic on HN right now is not the nerds. The entrepreneurs and grifters are desperate to find footing somewhere. They continue to double down on the alleged impact of "AI" and are now asking that their followers remember the moments they are most nostalgic for, assume everyone else was doing what they were back then (fumbling and confused), and ignore all the rest of history.
I doubt you ever understood the solid state physics, semiconductor fabrication processes, supply chain logistics, monetary policy, shipping routes, mining engineering, etc. "Knowing how things work" is a stone-age attitude.
Obviously no one knows every detail all the way down the stack. It's not possible, but having some level of understanding of how a system operates is useful especially if you ever have to fix it yourself or explain why/how it is broken.
Every domain you've enumerated is but a niche. Computing domain is a behemoth next to other domains, since computing itself is at the center of it all when the software is eating the world. Yes, they have their importance, but they are also mere bricks that built the current state of the world, next to computing becoming the mortar that holds them together. Dismissing computer illiteracy as inconsequential, (or at least comparing its importance to something like "shipping routes") in a world where computing systems are (in one way or another) everywhere, it's... just puzzling!
I do at least try to understand all of those things.
Obviously it's impossible to understand the whole stack. My rule of thumb is to understand at least two layers of abstraction above and below the thing I'm actively working on.
> "Knowing how things work" is a stone-age attitude.
Should we petition to rename this site to "Stone Age News" or something then?
Maybe I'm wrong but I always thought figuring out how things work was pretty core to the Hacker mentality.
If being a Luddite means I'm the only one who cares to understand anything, I'll accept the label gladly.
You: I think it’s good to understand some things.
This guy: If you don’t understand everything then you might as well not understand anything!
Guys like this are a corporation’s wet dream. Total intellectual dependence.
It's not hard to understand the physics and fabrication.
The rest of that isn't part of how a computer works.
There is plenty to learn about hardware / integrated circuits.
> "Knowing how things work" is a stone-age attitude.
What absolute nonsense. To openly admit to be proud of one's ignorance.
> The graybeards are aging out, nobody compiles their kernel anymore, and someday something deep will break and there will be no one left who can climb down and fix it. Maybe. But I think competence is the part that’s fine.
> knew a beige computer in 1995 that wouldn’t run a game until I had rearranged its bits by hand. More dependent than ever
If you look at previous article from this author, it says how Mac is amazing and how Linux sucks. Kids like that in 1990ties would buy expensive consoles, and would not deal with hack PC's to get free games.
Many people today are still dealing with cheap shitty hardware, 7 years old Android phones and sketchy ROMs... Just because there is no other option!
https://unix.foo/posts/it-will-never-be-the-year-of-the-linu...
That other post is just plain wrong. LLMs are just about the only thing that can meaningfully benefit from having access to almost all the code your environment is built from. If you're on Linux, the agents not only have all that code accessible in web search, but also almost certainly include most of it in their training data.
If anything, agents are an opportunity for the "Linux year on the desktop". Granted, I don't personally believe Linux needs to have that year; it's just fine without any mass adoption there. It's simply no longer that important a segment. I thought most of the debate around that also died out years ago (perhaps even a decade?)
Still, I recently had the opportunity to compare how Codex configures macOS and Fedora for the same task, and the latter was a bit faster and significantly cheaper (in tokens). That's obviously anecdata - maybe the particular task was overrepresented on one side or something - but I wouldn't be surprised if this was a more general trend.
>The modem sang its negotiation out loud, and that shriek was two machines arguing about how to speak to each other, and after enough times you could hear when the argument was going south and the call was about to drop. You set little jumpers on drives with your fingernail. You knew which interrupt your sound card answered to, because if you guessed wrong it answered to nothing. The machine was made of edges and you cut yourself on them, and that is how you found out where they were.
Wow. I had literally all of these experiences. I remember how you used to have to remember the Sound Blaster interrupt settings and pass them to any game that used it. And I'd completely forgotten about IDE jumpers, but yep - how fiddly it was when you wanted to put 4 hard drives in the same computer, figure out which were slaves and masters, etc etc.
I wanted things to be a little easier, but not this easy
I think back to this 2019 talk about software engineering by Jonathan Blow, entitled ”Preventing the Collapse of Civilization”, more and more often lately: https://www.youtube.com/watch?v=ZSRHeXYDLko
>The knowledge is not in danger, in fact, it has never been safer. The AI models have read every manual that no human reads. They will recite, flawlessly and forever, exactly how all machines work. If this were only about competence, it would be the most secure moment in the history of computing.
Obviously this person doesn't have enough experience with the current crop of AIs. This is the exact reason why we should be concerned in the first place. Someone has to design and build the next version of everything, and AI isn't qualified to do the job, and might never be.
>What is dying is acquaintance. The plain, unglamorous intimacy of having fought a particular machine, and lost, and gone back, and finally felt the thing give.
This is true. But then again, who among us had to get up in the middle of the night, and trudge out to an outhouse? Who had to shovel a bit more coal into the furnace? Who had to pump water, and heat it on the stove? Who had to split wood, and stack it for the winter?
There are many, many things we're no longer acquainted with, and that's sad, but still ok. What's not ok is losing the competence required to maintaining the infrastructure and supply chain supporting society and civilization.
I came into the US tech industry in 2011 from a CS program (programming on my own longer), which was basically a pipeline into Java or web dev at that time (the classes on other things like OS, compilers, and FPGA were a thing, but seen by most as a hurdle not a promising or practical career pathways).
Most of the people I dealt with for the decade following that (i.e. pre-LLM) had little clue how any part of computers worked, and I don't intend that to be particularly pejorative, it's just that it's been somewhat easy to carve out a niche doing something to do with programming while having little understanding of the small or big picture.. all heuristics and applying working patterns.. for a few decades.
You can still understand whatever you want, today, with added tools. It's just a choice to turn your brain on or off. I think LLMs are perfectly fine as a learning tool to interrogate a subject, do comparatives, and then formalize your understanding by reading the sources. At a macro level brain rot is real, but it cuts across all generations and it started long before LLMs.
> At a macro level brain rot is real, but it cuts across all generations and it started long before LLMs.
The thing is, if there’s a way to do something easier people will generally just do that. All the nostalgia of the olden days has as a component that there were no alternatives as the example in the article highlights.
We can try to preserve the hard way to things that gets improved, but the hard way will only ever be niche if it doesn’t become extinct entirely. There is no going back.
Sure, humans are attracted to laziness by default and at its best technology must be used to lower toil and suffering (unfortunately there are unintended consequences etc for philosophical debates).
The bigger issue with brain rot is attention spans. Social media has ruined attention spans.
This might just be a personality trait of me, but I really like technology that's transparent enough that I can fix/modify/understand it. Opaque technology gives me an ick. It's not really a complexity thing -- Linux has this property of being transparent even though it's extremely complex. The way he describes old computers.. man, I loved that about them, there was nothing _secret_ going on, you weren't worried about telemetry or weird background processes. Then again, I still drive a manual transmission internal combustion car in 2026, so maybe I don't represent the world.
Then again, maybe there's light at the end of the tunnel. Even though AI boosters kind of annoy me, I do appreciate that they're tinkering and hacking and working to understand the technology on some level. I think there's always going to be some small number of people that are like this, in the same way there are still people that look at disassembly and understand CPU caches.
Two thoughts:
1: This is why I prefer console games. I just want to have fun without fighting with the machine.
2: There are plenty of people who appreciate old techniques and methods; and keep them alive. Think of going to a museum and seeing someone demonstrate an old craft or reenact how a craftsman did their job. For example, in my town there is an old, water-powered corn mill that still runs and sells corn meal.
At this point, I sound like a brand ambassador. But I highly recommend The Machine Stops by E.M. Forster. It's a story about what happens when the machine of modern life that we all rely on suddenly stops. It's been operational for hundreds of years, and there is no one left who knows how to fix it.
I worry less about whether people know how it all works.
I worry more about whether people care and consider it a problem when nobody knows.
The thing that freaks me out about "the end of programmers" is that we're going to introduce a knowledge gap. And I'm unaware of any knowledge gaps in anything else we do. For any technology we use, we have humans who can work together to build it from dirt.
When no one knows how to program the tools that we rely on to fly our passenger planes, drive our cars, and manage our bank accounts, what happens when the LLM gets it wrong?
I wouldn't calling it learning more and more like "cycling through SoundBlaster DMA and IRQ options until the sound work". Still, there was an intrinsic curiosity that isn't as prevalent.
>To play a computer game in in the 1990s, you first had to understand how the computer worked.
As someone who played plenty of computer games as a teen in the 90s I assure you that I did not know how it worked. I learned how it worked later when I got a CS degree at university. Following the instructions from the booklet that came with the game may have helped me to get it working, but there was minimal understanding.
This is sort of the story of the telephone system of the 1950s-1970s, or electricity in the early 1900s, or cars from 1950-1980s, or airplanes from 1910-1939.
I have no idea how an electrical transformer works (well, other than the bare theory I learned in physics courses), or how power gets from the power company to my house, or how the circuits in my home are setup. I plug something in, and it works, and occasionally I throw a breaker if something is malfunctioning. There's no resistance there (pun not intended), and there shouldn't be. People got killed trying to wire their own homes.
I used to read about phone phreaks from the 1970s that could do black magic to get free long-distance phone calls. When I grew up in the 80s, that was basically gone. You picked up the phone, got a dial-tone, and called. And now it's really gone, with everyone having an encrypted cell phone connection over 5G, and your IMEI and IMSI being phoned home to every tower you connect to.
It's the nature of technology and capitalism. As the technology matures, it gets hidden away to become increasingly invisible to the end user, so you just do what you want to do with it. And then the engineering resources get spent on new problems.
> And now it's really gone, with everyone having an encrypted cell phone connection over 5G, and your IMEI and IMSI being phoned home to every tower you connect to.
people will hack enterprise pbx systems. The stream just flowed somewhere else.
>What is dying is acquaintance. The plain, unglamorous intimacy of having fought a particular machine, and lost, and gone back, and finally felt the thing give.
I think this is a universal feeling that accompanies any technological innovation. My phrasing is that new technology unbundles the thing people want from the craft that was formerly required to get it; any craft requires someone to learn and achieve through struggle.
Yes, it was not easy to install and play a game on the computer in the 90's. Today, we also don't want to wait for the game to boot.
Speak for yourself but I certainly don't know how it works.
Also, if you don't think the difficulty is the knowledge, I'm sorry but I must disagree. There has been more than one government or corporation or large institution that has tried to destroy history and knowledge in the past.
I think it's fair to say that knowledge is also at risk.
I can't help but think this type of step function change has happened with computers before.
The change from punch cards to magnetic storage certainly made it so you didn't have to "know how it works" for every single bit.
The change from machine code to a language like Fortran brought about such an abstraction that a Fortran dev didn't "know how it works" at that same level anymore.
At this point, the layers of abstractions between using a React component and something being rendered is immense. React VDOM, the real DOM, browser render engine (which sits on abstractions like ANGLE, skia, etc), calls OS APIs, which call driver APIs, and the lowest level of anything it is still C/C++ that is compiled (abstracted) to something closer to what the hardware expects.
I won't bore you with the ChatGPT output details, but it estimated at least 35 meaningful layers of abstraction between a React component and being rendered to the screen. LLMs seem to be the latest level of abstraction to make it so we "know how it works" less than before.
It's hard to articulate why, but 35 layers of deterministic abstraction feels very different from 1 layer of nondeterministic one.
In particular, I think every developer has experienced the need to jump "down the stack" to debug or understand something (even if not all the way down). Certainly, I think any senior developer should be at least conversant in the first few levels below wherever they "live". But this seemingly ends up looking fundamentally different in the interaction mode of an LLM, because you'd just ask it to jump down the stack for you
I still see this more like there will be "devs" that only know how to use LLMs and not any layer under it.
Just like there were web devs that only knew jQuery but not any actual JS. Game devs that know only Unreal Engine or Unity but nothing about DirectX, OpenGL, or Vulkan. C/C++ devs that know nothing about LLVM IR or the actual bits their compiler generates. It can also be a poor understanding of the layers in the abstractions above you too.
I find a lot of value in being able to understand levels beneath the layer I'm mostly working in. There are things, even/especially with LLMs, that I can grasp because of my deeper understanding.
But there have been plenty in our field that have only known their level of the abstraction for quite a long time. Every time there is something introduced that lowers the barrier to entry to making things, there is a version of this "but they don't know how it really works..."
> The machine was made of edges and you cut yourself on them, and that is how you found out where they were
> ...having fought a particular machine, and lost, and gone back, and finally felt the thing give
mayhaps this is because our computing paradigms are stuck in the 70's
I'm still nostalgic for the times when I had to enter the bootloader using the keys on front panel of the PDP-11 to then loaded the OS from paper tape. (not to mention multi pass compilers using paper tape between the phases...)
With how seriously rise of LLMs damages schools and university education, and thus next generation of comp sci graduates, at least we can hope to secure our arcane knowledge jobs for long.
This is also a fear I have, I am answering it with building bigger things with AI: what I loose in expertise in the details, I gain in the pleasure of architecturing bigger things.
We always were the only people who ever knew how it worked. In 1990 people fellow students called me to fix their computer, they had absolutely no idea how any of this worked. No. Idea. Yes, the machine was being difficult; but their reaction wasn't to fight it, or understand it. It was to call someone to do it in their stead.
I'm not sure things are very different now.
Maybe the difference is more of the professionals in the field now haven't built that same muscle, as there's a broader group of people working in tech. Whereas the folks that could fix things in the 90s mainly gravitated to computers as a profession. Just random musing though I truly don't now.
I feel that things are pretty different. Although the example of interrupts and jumper settings being common knowledge is a bit of a stretch... it's still amazing to recall that MS-DOS was regarded as an end-user / consumer OS, and that, more generally, it really was regarded as totally normal to need to invest some time in learning about the system, files, directories, typing, configuring/customizing settings and network options just to be able to do what you wanted to do.
I find the current expectations around consumer "apps" to be totally infantile in comparison, where everything is now a single-purpose "app" that does exactly one thing when you push a button, and if you want something even a tiny bit different.. you can't, and that even basic things like files and settings are no longer accessible.
Compare with what the drivers/operators of the earliest automobiles were expected to know vs. what the same category of people are expected to know today.
There's nothing new about this particular progression - we've been through it in dozens of technologies already.
I think the difference is that with those technologies, not knowing is seen as a hindrance (I imagine most people wished they knew how their engine worked, that's useful) that one can simply live with, whereas with computers nowadays I see people just not care about these skills. When I do something interesting on a computer that other people see, the response isn't "huh, that's a neat skill" it's "why?". People don't care anymore. They don't see knowledge as useful or something that is beneficial to pursue
The graybeards are aging out, nobody compiles their kernel anymore, and someday something deep will break and there will be no one left who can climb down and fix it.
I dunno. I suspect it'll be slightly different in the OSS world[1]. There will be some folks who can climb down there and fix it, but since that skill is no longer valued (as in, being able to do that isn't valued by most users now, much less in some hypothetical future). And they will be expected to do that, now, without thanks. And when they say "I'm too old for this shit...you can go pound sand" because this was one unpaid insult too many, users will shrug their shoulders and load windows or buy a mac because those companies are paying & tasking people to keep things running. And by inches oss ceases to be a thing.
[1] Yes, I know there are many oss devs who get paid for their work. Many critical contributors do not.
> They will have a tool that does everything and asks for nothing, and they will be as easy with it as you are with the light switch you never once thought about.
Oh come on! Of course I wondered how a light switch work, and I then learnt. I remember taking apart broken electronics as a kid, and that later morphed into also trying to repair things. Including the computer (both hardware and software wise). I remember ending up reinstalling the OS so many times as a kid on the computer I had access to when I broke it in various ways past what I was able to fix.
Sure, not everyone will have that drive to understand how the world around them works under the hood. But for engineers and scientists I would expect a far higher percentage to have that sort of personality.
It doesn't matter if it isn't "pushing back", just that I don't understand it is enough to catch my interest and a reason to go poking at the thing.
The thing is, yes it will still happen, but at substantially decreased rate. Groups of people banding together to solve a problem or figure a solution will slowly stop forming. Everyone will be out there learning just for themselves.
The collective knowledge of people will likely ebb. Things will be different and individuals will have to adapt.
There is a lot of useful information on how things work out there, if you want to learn. YouTube videos took the place of popular science magazines largely, and perhaps it will be some different medium tomorrow. But the knowledge is there, and just like in the past you have to go and look for it, just in different places.
There are also still courses you can take to learn practical skills, I for example took a short blacksmithing course a few years ago. A lot of fun and way harder than it looks (at least if you want to get good at it, my goal was just to give it a try and have fun, and 5 evenings was enough for that).
Why do you have to know how the whole thing works?
You can know how your entire farm’s water system works. But a city?
Somewhere in between those extremes, you just have to let go.
Gee, I sure hope pump 6 never fails.
Folks who keep mentioning that this is no different than any previous upward leap of abstraction in human history are missing a key point that the frontier labs are certainly not missing: this is the first time in computing that you are becoming completely dependent on a _subscription service_. I don't need to know how my CPU works because it continues to work once built. Once I outsource all cognition to a billable service, I am forever and continuously in thrall to someone else's revenue strategy.
But this too shall pass, it used to be you were dependent on a subscription to just have access to a computer itself (mainframe) and now we all have one in our pockets. I’m sure some day we will be able to run something as good as Claude at home.
Some of the best systems programmers are under 20 years old what do you mean
"To play a computer game in in the 1990s, you first had to understand how the computer worked.
So you learned. You opened files like autoexec.bat and you read them."
Ehh I dunno about that. I rarely, if ever, had to mess with any of that junk after Windows 3... I also didn't have to deal with any IRQ issues. So seems like it was already mostly abstracted in the "1990s" lol
Yeah same. I remember playing a bunch of Sierra games on my first PC as a kid on either. Most I had to deal with was installing the drivers for my SoundBlaster card from the included disk. Most I dealt with was putting in the CD, double clicking the installer and entering the product key.
That said I did run into my fair share of other problems, and that early era of personal computing and my access to machines is the only reason I work in computing/tech today. If my childhood wasn't full of tinkering with these fascinating machines, and I only ever had an iPhone or iPad, I likely would have turned out much different.
Anecdotally, I think it depends on age and family more than anything. I grew up with hand-me-down computers, games because we couldn't afford anything new. Started with win98 and games from the mid 90s, but running them in 2002.
I do remember having to look lots of things up and figuring out why some things wouldnt work. Then getting into building our own computers (because it was cheaper) and figuring out how to get halflife mods working...
If you wanted to play a PC game in the Windows 3 era, you were playing a DOS game. Playing it inside of windows just made it even more difficult.
Windows 3.x did not cohabitate well with certain games that required nearly the entire 640k of conventional memory to run. These are mainly games from the old DOS days; games that relied on DPMI were largely exempt. But to play certain Sierra titles, for instance, you had to set up one CONFIG.SYS configuration that loaded HIMEM.SYS and EMM386.EXE for Windows and another that left them out to free up that memory for your games. In later versions of DOS you could set up a menu to select the appropriate configuration at boot.
And of course IRQ diddling was still necessary to configure sound, network, and game controller hardware throughout the DOS era of gaming, which lasted well into the Windows 9x era.
Also, the quote didn't really apply in the 80s. Getting a game running on a C64, Spectrum, Amiga, ST or Apple II didn't involve understanding the computer.
That autoexec crap was unique to DOS PCs. DOS was a uniquely crappy experience in a lot of different ways.
https://en.wikipedia.org/wiki/When_the_Bough_Breaks_(Star_Tr...
Amazing how we always have a Trek episode for everything.
probably, simply not. People be working on physical things managed by AI!
Moral panic. We all have private tutors on every subject at our disposal now. It takes some special kind of mental gymnastics to conclude from this that no one will learn anything anymore henceforth. The article is just a long-winded way of saying "Kids today have it too easy", or, equivalently, "I'm getting old".
Having all the world's knowledge at our fingertips didn't seem to make us any smarter (quite the contrary, from where I'm sitting), so I'm skeptical that having private tutors at our disposal is going to. Perhaps for the ever-curious few, but for most people those "tutors" will probably just end up as another form of entertainment.
Yeah, I'm learning more than I have ever learned before. I suppose the worry is that people won't want to learn, though. That people will prompt the AI to just do things without caring about why and how they are done. That's a legitimate worry, but it's not new.
Desire to learn is deeply ingrained in humans and hard to root out. The education system does a pretty good job at it, unfortunately, but nothing ever is 100% effective.
I may not be a "typical" engineer as much anymore, in that I don't seem to be excited by knowing the intricate details of how things work as much as I enjoy knowing the capabilities of the systems that enable the projects I work on, and working at a higher level that might be more abstract and conceptual. AI enables that very well in that it enables jumping between paradigms and enabling creativity by removing a lot of the grind that was necessary in gathering technical details. I find it less appealing now to get into the weeds of how a thing works, even though at 10 I very much did. Like the author, I dug into the details of exactly which bits needed to be activated to make a particular program work. But as I aged, I was drawn to higher levels of abstraction, I think because I realized I could do more overall by piecing together bigger blocks. Yet I know colleagues who like to be involved in a very granular level of detail. I have many friends who almost revel in their knowledge that they are the ones who know precisely why something works the way it does. I think about the story of the NASA engineer on the Apollo 11 program who understood the assembly so well that he could confidently answer on the spot the moment the radar overload "1201 alarm" error flashed that the mission was still good to go. Some folks get so deep in this perspective that it almost seems like an affront to suggest they may be wrong about one of those details. And well they should, because precise people are very much needed to understand intricate systems. But it makes me wonder how they feel with AI being an indeterminate system - one where you can never really know what the answer will be every time. Likely still correct, and getting better every day, but also possibly different every time. Added to that, it's exponential spread across the software engineering domain. For me, even as someone who likes high-level systems, I still relate to the feeling that the detailed part of me that still likes some of the internals may be less focused on in the future. Even for me that's kind of sad. But I am also curious about my colleagues who are super detail oriented - how do you feel about AI and how it changes the focus on that detail aspect?
When I was young I was so naïve. The whole world seemed to be barreling straight towards the things I liked and was interested in. Every year there were more and more nerds, more people with computers, more people playing video games. It felt so much like I was the future, that in a couple of decades everyone would be chit-chatting about what was coming, buying and comparing devices, etc. I could not have conceived that people younger than me would have less computer knowledge than me. If at twenty you had told me that I would wake up one day and teenagers no longer knew what a file extension is I would've asked you who won World War 3.
I've been thinking that there might not be new programming languages. The amount of code in the current popular ones will explode, so that's what all LLMs will be trained on.
Good luck coming up with a new language and getting enough content out there that LLMs will be fluent with it.
If true, I think that's a shame. There's plenty of innovation still to be done.
How pessimistic you are on this subject depends on how much you believe in domain transfer for these tools. No programming language sprung out of a vacuum, and it seems reasonable to think that a LLM sufficiently trained on all the C++, Haskell, and OCaml in the world could probably do a decent job at writing Rust, even if it had never seen it before, given the rust compiler as a tool and rust documentation as input
It certainly does worry me, though. As does the increasing amount of materials and manuals that seem to be written assuming the LLMs will be the (only? primary?) audience
> How pessimistic you are on this subject depends on how much you believe in domain transfer for these tools.
I tried to force various LLMs to generate many strange languages for the past 3 years. From Raku to Smalltalk, to Common Lisp and Janet, to Prolog, D, Nim, Zig, Elixir, F#, etc.
It's a mess.
They do a mostly good job with language semantics. Agents can also use standard tooling to catch and fix syntax errors. So far so good, though if a given language provides an unusual implementation for a popular feature, the model can get confused. It's usually easy to spot.
The problem is with the ecosystems. They tend to be tiny, and for new languages, they will be tiny almost by definition. As a result, the vast majority of the generated code either reimplements things from scratch or hallucinates dependencies. Since the agents iterate and the hallucinated deps don't materialize, the model fixes that... more often than not by writing from scratch. Or leaving a `// TODO:` comment.
What I found helpful is straight-up vendoring deps as submodules. Armed with grep and sources, they are much more likely to produce working code. But that also means you still need to handle the dependencies yourself, while it'll "just work" for a Kotlin/Gradle project.
It doesn't matter for all projects to the same degree, of course, but it is a pain point that will add to the difficulty (sky-high already) of a new language getting any traction, unfortunately :(
This has always been true. I never fixed my car. I knew how it worked well enough to know, hey, that sounds like its coming from the exhaust pipe. Then I took it to the mechanic. I can do basic maintenance on my bike, but I still take it to the bike shop. I have a small collection of vintage cameras, which means tracking down the few people left who know how some particular model works, might have parts. If your Synchro-Compur shutter needs parts, forget it. For most people, most of the time, the assumption has always been that someone else knows how to do that.
It doesn’t have to be this way, but the cost is performance (and falling behind competitors).
> The knowledge is not in danger, in fact, it has never been safer. The AI models have read every manual that no human reads. They will recite, flawlessly and forever, exactly how all machines work.
That's wrong, and that's exactly why the loss of knowledge is such a problem. LLMs do not, and cannot, actually know a single thing. They are a statistical model, not knowledge. When they give out wrong information (and they always will, by their very nature), you need someone with actual knowledge to be able to recognize the BS and correct it. But we are losing the knowledge, and unless things change we will be no better off than the people in dystopian sci-fi stories who pray to the machine god because nobody knows how it actually works.
I can already see future where there is group of people goading AI to right direction by repeatedly changing what they ask slightly. And then memorise the times when you got right incantation for seemingly right outcome. Without actually understanding or being able to reason about process in the middle. Possibly a seemingly inconsequential misspoken word or typo can lead to better or worse outcome. Or maybe just not saying "Please" will sometimes produce wrong output. and other times you must not use it...
Sounds like absolutely horrifying dystopia.
Communing with the machine becomes spiritual. Computer operators will become a priesthood.
Welcome to the future. Make no mistakes.
Yes. And since LLMs can not improve knowledge - I mean, they can generate new arrangements of information, but they have no idea whether any of it real or making sense, unless humans - explicitly or through training - tell them, the more we rely on LLM knowledge the less the quality of it would be. Right now the LLMs are mainly in auxiliary role, so most of the knowledge erosion they generate is laughed at and relatively quickly corrected. But would this hold once the role of generative AIs increases? We already essentially entered the chaos period with news content - there's so much noise that it's basically impossible to know if any news message you read is true or manipulated somehow. This is going to start happening to more fundamental knowledge too, either on purpose or just by the force of the probabilistic nature of generative AI.
I was just thinking. What is the future when LLM is used for both code running itself and then also designing the hardware it runs on? Will it be sustainable? Or will there be some sort of error cascade that destroys the whole thing?
Maybe AGI is impossible with current model as it simply can not reliably improve itself... Enough errors in any part of loop will stop the progression.
I think what you describe is called the Model Collapse.
> Yes. And since LLMs can not improve knowledge
1. GPT proved Erdős Unit Distance Conjecture entirely on its own
2. GPT-5.4 Pro Solves Erdős Problem #1196 (April 2026)
In fact there's a whole benchmark that's measuring this: https://epoch.ai/frontiermath
I'm sure this comment will get buried, but I wholeheartedly agree with your take.
Imagine if we had LLMs back then to write xorg.conf for you and it hallucinated the modeline and you blew up your monitor.
This comment might have gone hard in 2023. Now it seems out of place. LLM's do hold knowledge empirically. They sometimes give out wrong information, like humans. And you need someone with better knowledge to correct it.
> But we are losing the knowledge
No we aren't and this is spreading FUD. Things have always been like this. Its called specialisation and this is how society progresses. I don't know how the supply chain worked to get the food to my table. That's why its so cheap!
Paging Vernor Vinge to the white courtesy phone.
There are more important battles for humanity to fight than someone trying to figure out how to configure Linux correctly for their hardware. I won't miss the need to waste time fighting the computer going away.
Try getting a Linux laptop to just go to fucking sleep and not drain it's battery flat within a few hours, and wake up without losing wifi.
Just wait until they shred all the old useless books
I recently had a run-in with security restrictions at work that only blocked the abstraction layer at the top, and were utterly surprised to discover that if you go down one level none of their restrictions did anything.
They eventually decided to keep their head in the sand and pretend like their restriction was useful.
The issue was a complete lack of knowledge of how the system actually worked one layer down.
Does anyone have a list of all the times this sentiment has been incorrect over the decades?
There's no arguing down that we have a greater amount of information about old hardware and software easily available now compared to back then, and there are many more enthusiasts as well.
This kind of cognitive dissonance has been going on in other disciplines, like music, since their inception. Guess what? There's thousands of highly talented young people on social media covering it all.
What people are actually upset about is that there isn't new "old stuff" and they can't make money off what can now be thoroughly understood by anyone.
Don't get me wrong, AI is a total shitshow and we're entering another winter, but there's also no reason to be jealous of what is making money right now.
Who cares?
There have always been layers of abstraction. I've been around for a while, and when I was a kid, the two choices I remember seeing were assembly code and simple semantic languages like BASIC.
Assembly seemed like too cryptic for me to really even follow and I never really did learn it, but at the time I remember people would say that assembly was easy and basically plain English compared to machine code.
As recently as fifteen or twenty years ago, I would occasionally check in and think of how unbelievably far away we had gotten from how the computer actually works. Like, you can just write "open window" and a window opens. Amazing.
Of course, those people writing machine code didn't need to really understand what P and N were in a transistor, let alone how an integrated circuit pulls it all together. And I'm not sure how much those guys knew about silicon dioxide.
The more complex things get and the more layers of abstraction there are, the more impossible it gets to really master things all the way down to first principles.
So what? People can carve out whatever chunk of the stack they want to really understand if they want to focus their lives on it. And for everyone else who's just trying to accomplish some other goal with computers as the tool, they will naturally use the highest level of abstraction and the simplest one for them to use, which is exactly what they should do.
https://en.wikipedia.org/wiki/Profession_(novella)
Competence Does Not Mean What You Think it Means.
Please don't fucking tell people that AI is competent. It is not and it cannot ever be because it is not alive.
AI can't be a competent person, but inanmate things can be competent. https://www.merriam-webster.com/dictionary/competence
I disagree wholeheartedly. I do not give a fraction of a flying fuck what MW thinks competence is. I'm a goddamn engineer and I know what it means
If you use your own definition for words, don't be upset when there's confusion about what they mean when you try to communicate.
Sure, but don't think I'll be impressed by pedantry when there is no real doubt what I mean.
People nerd sniping the op add nothing of value to the discussion. Take a break from HN if you’re just going to shitpost.
This and it was demonstrated in Star Trek The Next Generation: Season 1 Episode 17 "When The Bough Breaks" [1]. They did not know how to fix their computer system or even where it resided. It did everything for them but was also killing them and making them sterile. There was nobody left to figure this out until the Enterprise crew happened along. This could happen on Earth. Who will save us?
[1] - https://www.youtube.com/watch?v=5WWyKQThSKg
>The difficulty was the knowledge.
Please write your blog posts yourself. I don't want to have to parse through this stupid claude output to get to your point. Literally just posting the prompt would be better than this.
We have very different perception of the same shared reality. I read the entire blog post and it didn't cross my mind that it was AI written.
That's not an AI text.
It's 100% AI Generated, stick it in Pangram yourself and see.
Hmmmm, I didn't read that as AI. What gives?
It reads like marketing speak is the problem. Why is it written in such a urgent, compelling manner with those short punchy sentences? It may not be AI, but no one writes like that naturally. Maybe the author picked it up from reading a bunch of AI writing? The repetition of using you as a subject is also a red flag.
It's the same style I see on Twitter and LinkedIn a lot.
Perhaps that's a you issue? but it has tells all throughout the text.
Would you mind elaborating on what other tells the text had that it's written by AI?
The top comment only based off of one sentence which to me doesn't seem AI. And the author is saying in comments here that it's not AI.
I am genuinely curious as I'd like to update my brain on how to detect AI slop if that's the case here.
> "they will be as easy with it as you are with the light switch you never once thought about."
This metaphor doesn't really feel coherent.
> "What is dying is acquaintance."
This is a very odd topic sentence.
> "The kids coming up will not feel this as a loss, and they are not wrong not to."
Claude loves stacking big piles of negatives and arguing both sides of everything.
> "I went looking the other night for a recording of a modem connecting. People keep them around like pressed flowers."
No they don't?? People don't keep recordings of modems connecting around like pressed flowers. I, of course, understand that this is a simile. The simile just does not make sense.
> "I played one back for no reason except to hear it."
Again, note the unnecessarily florid negative phrasing. It would be much more parsimonious to write "I played one back just to hear it"
> "I know that sound by heart. People will never know the machine that just played the sound back to me. Not the way I knew the one that used to make it."
> "The difficulty was the knowledge. You came to know that machine the way you come to know anything that pushes back. The resistance was the whole medium."
The X -was- the Y, twice in short succession, to say the same thing.
The problem here is, you might be right, it makes sense. But, make it more approachable. Tell people why you think that, elaborate. Don't assume everyone is on the same page.
My problem is that my personal AI detector running on wetware just goes ping and it's hard to articulate why. I agree with the person you're replying to on all points but also, when I think something is AI, it's very much just a little brain ping that I would have to laboriously try to work backwards from.
LLM text often just is missing the cadence of human-written text. It never slows down or speeds up; it keeps a relentless pace. Sure, a human can write like that, and I'm sure people do, but I find it enervating to read, whether it was a human or an AI that wrote it.
There's an old saying of "Never Wrestle with a Pig. You Both Get Dirty and the Pig Likes It" which I'm about to fall victim to here.
The bulk of your comment history seems to be finger-wagging these kinds of slop accusations on others. Maybe you're right with the others, but I definitely know you're not right on this one.
What you're pointing out here sounds more like you just don't like the style of writing and the attempt to add some nostalgia to it, which is perfectly okay to not like. That's not something I can do much about and I'm not forcing you to read the post.
Yet you clearly read the post, and I think we'd all get more out of hearing what you think of the underlying premise rather than you circling some adjectives.
What did you mean by "they will be as easy with it as you are with the light switch you never once thought about"?
What does it mean to be easy with a light switch?
A shame that Pangram flags this as AI.
https://www.pangram.com/history/c0a9cde2-7a5c-4588-83a3-0269...
Given that I’m the author, I promise you I’m a real human. It’s not AI.
Obviously there’s a rise in pure AI writing but I find these sort of services that “analyze” content more harmful than good.
“Pure AI” so you haven’t used any LLM at all, or just partially?
No one questioned that you're human, but the writing is very apparently AI-written.
Pretty much every HN article now has someone in the comment section chiming in to complain "This is AI".
Sometimes that's me. I wish more submissions didn't read like AI. I worry that people can't detect tiresomely LLMy output and upvote it to the front page. I'm not mad or anything, I just find the internet less and less useful, because so much that circulates pings my personal AI detector (running in my skull). Sometimes I just close the tab in frustration but sometimes I comment, because it sucks.
(I don't actually care that much if it is LLM text, I care when something reads like it is, because it means everyone writing blogs using this style sounds like everyone else. it's alienating!)
It is interesting that your recent posts have much more LLM style tics than your post on the same site from 2023. Are you deliberately imitating LLM writing quirks?
Why does it have to be deliberate? It's not surprising that people exposed to output from LLMs will unconsciously pick up their linguistic habits.
Maybe it isn't deliberate! I was asking out of curiosity. This reads as extremely LLM-y to me, so if it was purely human-written I am curious about the process.
Even more worrying if it's not deliberate and we are all just naturally writing like that now.
I trust these detectors even less than I trust LLMs in the first place.
The satirically brilliant Genius AI detector correctly indicates it is not AI
https://geniusaidetector.com/
I submitted a few of my papers from grad school (long before LLMs or so-called "AI" existed) and they were flagged as being ~70% AI.
I know you know this, but for those reading, these detectors are bullshit.
Damn, that is depressing. What a shitshow we've created.