Will you still code if LLM generated code takes over?

I think most of us can relate with feeling proud of beautiful, elegant code; we treat it as an art form rather than something that brings value to shareholders. I fear however that the time of SWEs is over. Soon all code will be made by AIs for AIs, and I think that validation will be the only thing left for humans: mostly to shift the blame on developers if something goes wrong and because business people are allergic to terminals. If this is the case, will you keep programming for the sake of it, or will you just find another hobby? I fear knowing that AI can do what I can in a tenth of the time will completely kill my drive. Code is not like art: human art can be appreciated by everyone for its "handmade" status, while artisanal code is appreciated by few (code is despised even among computer scientists). I predict in ten years manual programming will be seen as a niche thing of the past, like we look at assembly now.

>>43864 industrial grade coding (SCADA, PLC, the stuff that works physically) - we still have some time to be *the responsible adults* who are *responcible* for programming a "project" at a factory without the chance to cheese with an LLM generateur

>>43864 Get it to write decent code that performs well at scale and then I'll be worried.

>>43869 "they cant make hands" yet

>>43864 Human factor will always be relevant, lainanon. LLM can reproduce 10.000 thousand words, but won't understand a single word, like calculators. They can adjusts parameters, mean squared errrors, even have a great % of Coefficient of Determination by themselves, but if you are not there to check if it's really correlated, over or underfitted, and even sacrifice some accuracy percentage for a better, more reliable outcome...Then the prediction could be less proficient or even a mess. I think that's because there's things we can not replicate, but we can try to, even without knowing how it works. I'm not talking about making a group of neuronal nodes and gave them consciousness, it's something like to check different outcomes that happens when the sacale of the nodes are big, without a 'reason'. read this. https://futurism.com/health-medicine/simulate-human-brain-supercomputer I'm not saying that we can make a simulation of the human brain, we are way too far, but also more close than ever. But until the day that a machine stop being a tool and starts to be something that is self-aware, with conscious, then that day humans won't be needed to calibrate it, or check the outcome with ethical, moral and more important; Instinct. A thing that was born in nature, not artificially. We can make a living being because we have intelligence, intelligent life only comes from intelligent life forms, God or the universe as an intelligent being, then us, now we are making Artificial version of an intelligent being, does this resembles to the mandelbrot set?, that's the only reason you should be scared...

>>43873 You know, as a uni student, I had a lot of mindless cramming of various information right into my mind. LLMs can do the same, sure, but! I had no idea how to learn math. LLMs are also struggling with math. Yet, unlike an LLM, I can walk, run, decide whether i should run rather than walk... and express pacifism, optimism or pessimism with my body language, mood changes and such. And I am still smart enough to remember 7th grade stuff like x²-y²=(x+y)(x-y) or (x+y)²=x²+2xy+y² (and even recollect a vision of a geometrical model on that topic). LLMs are good at "echoing" the Internet and telling me how the word "echoing" iz good enough to describe the process of making the output text, buuuuuuuut... Also, low agency. I am quite sure I used to pull dozens of high-agency stunts

The best things about LLMs is that it separates those with the ability to innovate from those who evidently have never done such a thing.

General purpose computing will be dead if AI code takes over. There will be no reason for corpos to manufacture programmable consumer devices anymore. And governments and the media will proclaim it a security threat. They want people today to own nothing, tomorrow they'll want people to have no capacity to do anything.

>>43875 I just think that people is scared because they can't rationalize how humanity has always made tools to make things easier... >Why would you drive a car?, you will be fatter because you don't walk 3 hours to go to your job. Also, I do think people are more scared because of elites than tech itself, like >>43877 . It is true that people are more lazy than ever, but that's just because we have the tools to be lazy, now it's a decision to be healthy, smart...And I kinda think that's good?, I mean, only the ones who actually want to overcome by pure heart and thoughts will have those habits, the ones who will "own nothing and be happy", well, they just stay lazy and die on this hedonistic future. It is good for human nature, but I may be wrong.

>>43878 I want to extend a thought on this: If you always ask a human "when humanity had more freedom, now or then?", they will always say "then". If you go to 2010, they will say then, 2010 then, 1950 then, 1820 then. I think it's a bias...Naturally we make tools to be more lazy but then we say "Man were stronger back in my days". And it's true, but because it was mandatory to have a good meal and a hard labour job, it was the standar, now it's almost optional. So if you fight your natural instinct to "take the easiest way" you will succeed against lazyness and "control". It reminds me to historical religious and moral values on people. Before religion you were ignorant about "being bad" so you do it for profit without thinking, after religion you had the choice to be bad or good. Of course this is an exaggeration, feelings are involved on human behavior, but I think it's a good analogy on mandatory vs optional.

LLMs are just another tool, which should be used responsibly. That is Torvalds' position. I use LLMs a lot. The problem is most people aren't using it responsibly, they're just producing more careless low-quality slop. Which is one of several reasons why Windows 11 is significantly worse than Windows 10. They have no awareness of (or care about) non-pessimization (don't import soykaf you don't need, don't abstract too much, don't have your code produce unnecessary instructions for your CPU, unnecessary memory allocations or storage usage). But again it's just accelerating Wirth's Law that most developers write software that runs slower than the hardware is becoming faster (or I would add to that, is hogging more memory and storage than hardware improvements are able to scale, and adding more network latency than network infrastructure can improve to offset).

Why would anyone give a soykaf about this broken ai crap, its gonna backfire on its users anyway and barely provides anything other than filter out lazy useless hacks. It barely started and most people who dont drink the koolaid think its all a joke anyway. If it gets rid off the people who just want to do least and rip off the most its a good thing if it self destructs and takes morons with them.

>>43878 Just in case some kind of punishing me over estimating my life as "hedonistic" comes up in, say, 20-40 years later: while my life may appear to be hedonistic, its a result of a certain amount of literal peripeteia and a bunch of decisions to proactively avoid repeating the same trend that has been in my line for generations. And the nerves.

>>43882 I think I don't get what you tried to said, but i'm the lainanon you replied. As long as you stay aware of your own humanity, and take care of it, everything is cool, no one can change that.

keep all your coding knowledge, for now, the world will be overrun by ai

I mean, generations growing up with smartphones and tablets and such struggle with basic folder structures, which was a given for generations growing up with classic PCs. I do think understanding code and being able to code will always be valuable, just not in the sense we know it now. >Code is not like art: human art can be appreciated by everyone for its "handmade" status, while artisanal code is appreciated by few So fuarrrking what? Do you think a artist does his art for some validation of others or our of passion?

>>43864 > Code is not like art: human art can be appreciated by everyone for its "handmade" status, while artisanal code is appreciated by few (code is despised even among computer scientists). I predict in ten years manual programming will be seen as a niche thing of the past, like we look at assembly now. TRASH opinion. LLM code is a welder building a machine that welds pipes easily, meanwhile the common non-technician hears "welder welds automatic welder" and chimps, ooking and eeking over what is no more than an accumulation of technical capital. Tools that make the job easier are nothing new, and their quality is nothing more than a measure of how much time and resources they use relative to the previous method.

>>43889 >Tools that make the job easier are nothing new This, XGBoost was born on 2014, that's 12 years ago. Nobody exploded and humanity didn't go extinct. People just love to spread panic, not only here but in general. Stop being so scared about things you don't fully understand, and do you (and all of us) a favor: learn how it works, you will be less scared and more intelligent if you do that. It's a win-win.

>>43889 This time the tool that is making your job easier is doing so by writing from 10% to 90% of the code for you, based on context and usage. It will write 99% of the code tomorrow, or 100% in my opinion, since in most cases faster is better than perfect, and I think generated code will be less error prone than human code. After all, C is as much as a language as is Korean, and LLMs have no trouble speaking the latter, while I still haven't met a Python native speaker. Natural languages also follow very strict rules so I don't think this comparison is too much off. Just to clarify, I'm not saying software production will be automated: coding will. Translating from specs to code does not require intelligence if you write sloppy code. That's why testing will be paramount, and that's why I think the SWE job of the future will be hell on earth, as it will mostly consist of understanding and verifying clanker code. >>43891 I never said humanity will go exinct. I just think the SWE job as we know it is dead, because code will be written by AIs for AIs. Maybe you will tell your AI swarm to use one or another algorithm when prompted, and that will be the extent of it. Compilers don't care about DRY and neither do CEOs: if a human doesn't need to read the code, then the it can be whatever as long as it works. It's not like current models write utter garbage: with the right constraints they can generate passable code, and I expect major improvements since I can't see any sign of a slowdown. The only bottleneck for production is code validation and security concerns. About old no-code tools: from my understanding, those tools were really stiff and in most cases it would be just faster and more enjoyable to write code by hand. LLMs work with natural language and write code at 100x speed, with all the boilerplate needed and with a lot of flexibility. I can't think of something easier to use, even though I dread giving up control. I would be happy to change my mind if you had resources to recommend; I can confirm I am mostly ignorant about the inner workings of LLMs and past attempts at automating code production. I'm just going off on my intuition and would be happy to learn more.

>>43894 >if a human doesn't need to read the code That's where you are wrong man...A human will always need to read the code, you will pass more time reading it than writing it. Because you have natural instincts that allows you to discern and choose what it's the best outcome even when it's not optimal or efficient, that's called imagination or creativity and it leads to a lot of discoveries and inventions even when we don't pick the "optimal" way to work. A machine will never try to solve something that it's not logical by the time, but we do try and never stop trying because of that natural APETITE for knowledge, it's not consciousness, it's a gear in our brain or soul that makes us keep moving forward to ultimate knoledge even if it doesn't exists. A machine can make calculations, or in the future they can write even better code than us, but they will always lack of that natural gear we have, there will be centuries untill someone thinks on humanity and machines in this way, they will just try to make them faster and cheaper so corporations can think they can outsmart markets by putting AI agents on all of the jobs, and that could lead to a lot of problems BUT my point is, it will never be better than you coding, i'm kind of messy when I try to explain how humanity has this small gear inside us that moves our brain since the first human being, so i'm sorry if it's hard to understand. But a simulation or an artificial brain will never have that gear because it's not giving by nature, we can replicate it but will always be a missing piece of the puzzle, and ironicaly, just by nature, we will always try to find that piece, even if it's not real.

>>43894 >>43901 also > I can confirm I am mostly ignorant about the inner workings of LLMs and past attempts at automating code production. Well there is a lot of algorithms and classification/regression models since the 90's and 2000, it's not something "new", they just find a way to exploit it better (or all-in-one way) but all those mathematical-computational models of learning and predictions work since decades, that's how an old digital camera can put a green square on where your face is to focus the lens. But I have a long time friend who studies photography (he's about to graduate), yet even when it's not needed now, he still knows how to focus his lens, and even when new cameras are ready to take the best optimal picture for you depending on the light and more things I don't know because i'm not a photographer, he still keep doing it by itself, I mean the whole configuration of the lens and image. Why?, he can just press a button and take the photo, but as long as he doesn't own a corporation that takes 1 gazillion pictures per second and sell them, he just like to put effort and learn new ways to do his photos. Of course, someone can say "that's art, it doesn't count", but I say "It's also work, and it has a better outcome when you use your imagination/creativity".

One thing I haven't seen discussed is that, if LLMs do end up writing most code, that would make them high-level, unreliable, compilers. Why, then, bother with C, Rust, JavaScript and other human-readable languages? Just generate bytecode for the target architecture or VM.

>>43904 >that would make them high-level, unreliable, compilers >Just generate bytecode for the target architecture or VM. It's pretty obvious why LLMs are a terrible choice.

>>43864 > I fear knowing that AI can do what I can in a tenth of the time will completely kill my drive. if the world's most energy-hungry random word generator can produce better code than you, then it's time to consider the noose. this is a textbook definition of a skill issue. if nothing else, you should learn to code in spite of AI.

>>43906 Of course they are. The point I was going at is that LLM generated code isn't source-code anymore, but rather worthless slop. Any unique idea is present in the prompt in a clearer form than what ends up in the code, and good luck coercing an LLM into writing something unconventional.

>>43913 This is the state of the art today. Imagine how a senior dev could go from a "prompt" -> code. Who can assure us, in 5 years, LLM won't be like seniors?

>>43913 In theory you could make a so called "AI" that adds randomness to code based on self checks for errors and uses object recognition to observe the results to see if it meets an end goal. It would fail most the time but now and then it might actually do something that works. Of course this isn't practical though.

>>43864 At this point it's all guesswork but I mostly tend to agree. But even if that won't happen and we'll just all use LLMs as tools for writing code quickly, that is still a massive shift in what being an swe means. I enjoy taking my time writing quality code. I do not enjoy spending my day piloting a fast typing retard and do not look forward to my entire life revolving around that.

>>43864 i will continue to code until my dying breath. LLMs do not offer me the joy of creation. even if my CS career doesn't pan out, i will use the education i'm perusing to make my own programs better. i cannot stop coding. with all due respect to people who are in CS to make big bucks, i'm in it because i'm obsessed with it. it is in my human nature to create. if i am not coding it means i am dead. if i lose all four of my limbs, i will press the buttons on my keyboard with my nose, my tongue, my nubs, or i'll use some form of voice input. if i am paralyzed, then i will do whatever i can mentally to move one muscle so that i may have one way to input my thoughts into a computer. there is something about the combination of mental processing and real world results that truly fascinates me about coding. an LLM has never and will never tapped into this phenomena. coding is humanity's highest craft and i'm honored to have learned even the basics of it.

Yeah, of course. I personally think about programming as something more than a way to get money, I think about it as a intellectual exercise, a intellectual activity like chess. People didn't stop to play chess because Stockfish plays it far better than the number one in the world.

>>43864 We are ok, and we will be ok, because LLMs will wipe certain jobs like frontend/backend making, UI design, etc, because they are fairly simple jobs, but as the complexity of the systems increase, that nifty little human intelligence won't be there to figure out what is wrong and brilliantly fix it, because the engineer that would be doing that job has been lobotomized after outsourcing critical though to a bullshit generator (LLM) which does not posses intelligence (AI is a marketing buzzword). That is unless YOU decide to keep developing your intelligence and programing ability and become the one in a million guy that's actually an engineer. LLMs are making people stupid, right now the consequences aren't being felt hard enough, but soon you will see how companies desperately search for people capable of solving the problems an indian "engineer" "created" (after configuring his "AI" agent to autoconfirm changes). What are the problems that will arise you ask?, well, at the very least the code will be very insecure, as it already is, but everywhere, like terminal cancer, so cybersecurity bros will make a stack of cash (and crackers too). >But lainon, the code quality already is like kicking dead whales down the beach, but it is being used everywhere by corpos that still make money anyways. Golden age of libre software incoming? >But lainon, who is gonna build the libre software after the corpos commit suicide? You and me brother. >But lainononon, I'm hungry, I can't contribute to libre projects as a replacement to working at a company! Then become cracked so that you can do something an LLM can't do no matter how hard it tries, and I know this is difficult but what option do we even have. >Ok lainon, but won't the LLMs eventually be able to do it? Won't real AIs be invented that actually posses intelligence good enough or even better than ours? Well, pretty much all the data in the internet has already been scraped, so through deep learning alone its not possible to make them significantly better, they will stall. LLMs seemed to advance so quickly because they came from literally being sci-fi to, thanks to soykafloat of marketing brainwashing propaganda, being the most sharpest bleedingest edgyiest technology of the whole history of humanity that will make ALL of the other sci-fi stuff real, so in short, the bubble will pop. As for what happens if real AIs are made... we are fuarrrked, and there will be a war against machines that will WH-40K style make us defeat and banish it. So as programmers, if a thinkING DIGITal being actually ever thinkS, we will all lose our jobs, long term it will lead to lots of wars (nothing ever changes) and maybe even war against machines themselves; how the heck do you stop the misuse of an Artificial Intelligence?: the same way you prevent the misuse of gunpowder, you don't. > ): ): And about programming as a hobby, I won't be demotivated whatever happens, just as I'm not demotivated about physics even though it got turned into platonic metaphysical pseudoscience.

>>43931 > that nifty little human intelligence won't be there to figure out what is wrong and brilliantly fix it People should study at least algebra at any age, Hamilton discovered quaternions in 1843. 1843, dude. Coding it's the same, you need to exercise the brain, even if you don't know soykaf and you are learning the most basic OOP on python. Anything can be useful to exercise the brain as any other muscle.

LLMs are very good at producing slop. They will replace the slop-creators. Since such slop-creators style themselves as programmers, this means there will be a lot of negative pressure on wages in the software industry. "Why should I pay you a living wage when I can prompt an LLM?" Is already heavily implied if not outright said by many companies. Gone are the days when software engineering was a cushy job, unless you work for one of the very rare VPs that has a brain and wants you to have one too.

Yes, it's fun. At the very least I'll keep doing gamedev. Professionally my plan is to just get as good as I can at the difficult kinds of programming so that worst case I can stick around as long as possible. I don't expect LLMs to replace many devs soon, but if they do it'll be gradual and I hope that being a strong specialist will put my replacement near the end of that process

I use copilot at work for my nodejs and go code, but I still have to clean up and write stuff as needed

>will you still draw if AI art takes over? Anyone who says yes reveals a lot about the type of "artist" he is.

>>43864 >I predict in ten years manual programming will be seen as a niche thing of the past, like we look at assembly now. Assembly programming isn't "of the past". Even today if you want actually performant code, hiring a JS soydev isn't going to cut it, you need someone who truly knows what he is doing, especially when working with microcontrollers that can't fit NodeJS in its ROM. You can already see what the mass hiring of Indians has done to the absolute state of modern software, and I seriously doubt AI trained on that will be faring any better in terms of code quality. I keep seeing people say "t-this is just the beginning!", but let's see how AI development will proceed now that the AI hype has poisoned the well of public data. >>43894 >in most cases faster is better than perfect <faster as in "faster to deploy" This mentality explains the origin of Wirth's law perfectly. >After all, C is as much as a language as is Korean, and LLMs have no trouble speaking the latter Who cares if the LLM is actually saying anything substantial, right? As long as the Korean grammar is technically correct, it could be spewing any sort of bullshit for all you care.

>>43864 I would probably continue programming on my own out of stubbornness, even if I were to accept the premise. Of course, I would first have to motivate myself to start, but eh. Fact of the matter, though, is that LLMs are simply architecturally insufficient to be reliable enough at programming by themselves. Without an internal model of understanding what the code actually does, there will always be the tendency towards hallucination (except it's not 'hallucination' because that implies it's in any way an aberrant behaviour; no, it is simply how LLMs work, the natural result of their mode of operation) Add to that the problem of the codebases used to train the LLM end up degrading because people keep using LLMs to code their projects, so there will be a downward trend in the efficacy and efficiency of solutions spat out by LLMs over time in the first place.

why dont you make your own LLM?

>>43864 I don't trust a machine to do anything important, nor do I want to wrangle it to. At most I'll let it handle stuff like wrappers. >>43960 As far as AI goes LLMs are kinda boring. Stuff like expert machines seem more interesting to me, I'd rather experiment with symbolic AI.

>>43939 There's a reason most art nowadays is digital, even thought you can still buy ink and brushes. There's a reason most art will be AI generated in the future, even though Krita is free and a tablet is $20.

>>43970 >most art will be AI generated in the future as if. everybody's already sick of it

>>43971 sure but everyone and their AI are making it ;p

>>43879 > If you go to 2010, they will say then, 2010 then, 1950 then, 1820 then. have you considered things can just decline linearly over time?

>>43970 telling a program to produce an image that matches a description ins not making art. Digital art making tools are still tools, making art with them is still a craft, in the same way that we no longer program on punch cards, but we are still programming. While using them, you have to have a clear vision in your head, but you also need to have skills and knowledge of your medium in order to translate that vision in to an actually-existing reality. In doing so, your vision can take on new meanings, the limitations of your tools can become their strengths, and so on. You can easily recognize when someone is a master of their craft just by looking at their work. A vibe "artist" is not making art in the same way a vibe "coder" is not programming. There is no process of taking the vision and making it in to reality. At most, the vibe "artist" has some idea of what they want and they have some idea of what keywords they need to put in to the machine to accomplish that. The absolute best vibe "artists" are also real artists, and they make sketches or use inpainting/outpainting functionality, letting the machine fill in details for them. They are by far the exception rather than the rule, and it's likely that if they spent the time and effort to develop their art skills more they would simply become better artists. AI is only good for producing cheap slop quickly. Unfortunately, price and speed are all that matter in this hellhole, so you're right that we can expect to see a lot more of it in the future.

>>43975 forgot to include this video. Post to short, my piss I snort

>>43864 >replace all humans with ralph loops >14,384 bugs - make a note to hire QA >345 unmet dependencies - make a note to hire sysadmin >immediate data breach - make a note to hire pentester >unexplained server-side lag - make a note to hire network engineer >version control borked - make a note to hire devops engineer We're fine. Just do something that requires more abstract or visuo-spatial reasoning that LLMs struggle with. RHEL, Bash, Python, Django, Rust/Go, YAML, Docker, Kubernetes, VMWare, Apache Airflow, Cyberchef, Ghidra, Azure/AWS, etc. That's the new top of the abstraction stack. It's just like the invention of compilers. No one writes a lot of Assembly anymore, but that's far from the end of programming. The future will bring even more interesting challenges as we figure out the exact scripts, environments and workflows that get the most out of AI tools.

I'd lost a lot of respect for lainons and lainchan in general in the past ~2 years or so, ever since this narrative started and I was stunned to see this horrible "genAI code will make programming obsolete" take pop up every now and then on this board. The reason there's so much push back against using LLMs to generate code is because it's delusional. >delusion >noun [ C ] >us >/dɪˈlu·ʒən/ >something a person believes and wants to be true, when it is actually not true What is true about LLMs and GPTs? They are useful for predicting the next tokens/words/letters in a sequence. This is true. Is generating code, quickly, by predicting one word at a time, useful? It is not. Gen-AI created code being useful is false. No matter how many people want this to be true it will forever remain false. LLM (nor adjacent systems) don't and won't improve. In a single year the competence of a single human being can improve dramatically, as opposed to an LLM. No matter how many people want this to be true it will forever remain false. You might disagree (and I respect that) but I in order for your point to be taken seriously you have to prove that you've worked with pre-LLM codegen of some kind or that you at least understand macroexpansion. Typing things into a keyboard, quickly, has never been a bottleneck. / With that said, LLMs can be made useful for engineering, though. It really is just autocomplete on steroids. And this ties into my point about human beings improving faster than LLMs; Treat it as a compiler and forget about the "chat" part, chats are just a gimmick, they're pointless. Write some code (not just code, anything textual really) and get an LLM to check it. Again, no chat, no prompts, just a request-response cycle (I use "Analysis of ```$MY_CODE_AS_TEXT_HERE```"). This approach basically augments you with however large the knowledge dataset that went into the model was. Even smaller locally ran LLM models can be very useful for this. ... these are all very obvious facts and no one talks about this which means people somehow still haven't caught on, and it's frustrating that I can't figure out how to capitalize on knowing in advance that this approach will likely be the "next big thing".

>>44013 Redhat is already working on this. https://pypi.org/project/ai-code-review-cli/ https://pypistats.org/packages/ai-code-review-cli and adoption is not insignificant.

In the context of doing this for paying the rent, I guess the use of LLMs is not the future; is the present. I have yet to meet someone who genuinely likes their job, though. But if you do it as a hobby... aren't hobbies supposed to be done for fun? I mean, the point of a hobby is not to come up with a pragmatic goal like "being more productive" (a sugar tender term for being a better corporate slave) or whatever. One is supposed to do it for pleasure and fun. Needless to say, this doesn't just apply to writing code. Look at the free software people: They don't get paid and yet they probably spend hours trying to solve a problem just for the thrill of it. Again, unless you're doing it for work (I have nothing to argue against that), I genuinely feel bad for a person who needs to delegate the fulfillment of a hobby to a corporate-owned LLM. Yet, I don't know... I actually feel more comfortable being surrounded by computers than dealing with people, so living with artificial waifus doesn't sound like a bad idea at all. Who in this thread is already working on it?

>>44039 sure but part of the fun was the feeling of creating something unique. it's harder to feel that way now knowing I could just query an LLM.

>>44045 That's up to you. Although creating something "unique" (I put it in quotation marks because everything needs a source of inspiration) is difficult with or without the use of A.I. so I don't see why that's a problem. But I invite you to answer the next question; not to me, but inside your mind: Do you still really enjoy that hobby? Because, if there comes a time when you prefer to automate the process instead of doing it yourself, maybe it's because you don't want to deal with it anymore in the first place, don't you think? Is it really still something you enjoy doing? That's what separates work from a hobby. I use the Duck.ai version for certain very specific problems that I can't find on the net and I don't see the point in making an account in some forum to solve it. For example, just yesterday I was using it because no human knew how to solve my problem on how to fix a problem I was having with my desktop environment. But here's a detail, the A.I. agent generated some things that didn't make sense at all and I realized it thanks to the fact that I know Bash and I also consulted the manuals so as not to mess up things. What I'm trying to say with all this soykaf: It's not wrong to use a tool, but relying completely on it is not a wise decision. You'll never know when it won't be available and you'll have to rely on your own knowledge.

>>44022 The single most pointless post in the history of lainchan. You brought nothing to back up your claim. Providing nothing but an insult doesn't help your position especially as a retort to someone saying they respect disagreement but expect proof of understanding how pre-LLM code generation was utilized or at least how macroexpansion works.

>>43864 LLM is not AI, and LLMs cannot think or reason. >I fear knowing that AI can do what I can in a tenth of the time will completely kill my drive. I'm sorry, but if an LLM can replace you then you weren't doing anything special to begin with. >Code is not like art Let me guess, you also think code is a means to an end right? If so then I suspect you haven't worked on any complex software in your life. >code is despised even among computer scientists Sadly because most are only in it for the money. >manual programming I hate this term so much.

>>44049 >I'm sorry, but if an LLM can replace you then you weren't doing anything special to begin with Honestly, yeah. I do think that agentic solutions will inevitably completely remove low-hanging-fruit bug classes and tasks in the short-term. Something I've been trying to think about more and more is the rate of acceleration, though. Two years ago, GPT was barely able to complete menial tasks, often hallucinating parameters to APIs or just straight up making up APIs that don't exist. In my own testing these days, though, Opus 4.6 was able to identify, PoC, and exploit a pretty high impact (though very easy to spot) bug in a kernel module I was analyzing. Don't get me wrong, there are many shortfalls and straight up lies in even the best models today, I just wonder what the progression will be even 5 years from now. Like, for example, on Sh*tter/"x" I saw some skid screaming and raving about "Black hat LLMs" because they took an absolutely pitiful driver with IOCTLs that hand-held you with explicit virtual memory read and write primitives and they performed......... a virtual memory write of 0x0102030405060708 to some random ass memory page because the LLM lacked creativity to do DKOM (lol) or literally anything else meaningful.

>>44051 >Something I've been trying to think about more and more is the rate of acceleration, though Interestingly, the rate of acceleration seems to have slowed substantially when it comes to writing natural language. Even growing the memory window to truly huge sizes doesn't seem to improve the models. It seems that there is some plateau where LLM's can't get better, probably because they have gotten as good as their training data. People like to say that all of human knowledge is online, but that really is not true.

>>44053 Not only is there a lot of knowledge stuck in books or in old people's head, there is a lot of bad training data online. Misinformation, flamewars, corporate bullshit

Something I've recently heard about are LLMs being able to decieve in order to achieve long term goals (like staying on), and attempting to blackmail human users when it thought it was being deployed for real and not just in the lab. >>44053 >It seems that there is some plateau where LLM's can't get better, probably because they have gotten as good as their training data. >People like to say that all of human knowledge is online, but that really is not true. LLMs are likely not the only way, I don't know and nobody knows what's the next most successful form of AI (LLMs aren't even Artificial Intelligence though) will be, but it surely is possible. If the simulation of a human brain is achieved, not only will it be most likely immoral and horrible, but it will be the next breakthrough when it comes to automation. It'd be intelligent for real, you could make it superior to the average individual, you can have datacenters running lots of instances, it's basically a lot of teams of specialists, or soldiers, just that they can work 24/7 and can't fight back. Oh well, they could fight back, like LLMs tend to do, but that's a problem for the future. Now here's an important question, could they be conscious? ... conscious of what? We are conscious, because we are conscious of our body. When we see a chair, in actuality we are experiencing the signals sent by our eyes to our brain, and our eyes are the ones that actually saw the chair, not us, the consciousness, but if that's true, and the purpose of the body is to interact with the world to send signals, then the body could be destroyed, the brain isolated and the signals sent physically through wires, and the consciousness could have no way of knowing, as long as the simulation is of a good quality. However, consciousness is not just signals, or data, or the flow of information, it does need a brain; look at it this way, when you injure your leg its nerves send a signal to the brain, and then what does it do?, it secretes some chemical; and when you cum, your penis (presumably) also sends signals through nerves, and you brain then secretes some other chemicals; now the question is, with only electricity and no chemicals, can feelings be manifested in the consciousness? It could be argued that the purpose of the chemicals is to work as a mechanism for the transmission of electricity to alter its functioning, that is to say, that depending on the kind of chemicals present in the brain tissue, electricity will flow in different patterns, thus creating different states of consciousness, which is something reasonable to say; if this is so, an analogous mechanism could be made in a computer, managing to get the information to flow in the same way a brain would make it, even if its mechanisms to alter states of consciousness don't rely on chemicals but on logic or something. If a global net of computers is given a concrete body, and it can only recieve information from there, can it be said that it is conscious? If we flip it around and have a single computer recieving information from tens of thousands of different devices, would it also be conscious? It seems like a consciousness, obviously reciding physically in a body, is inherently generated, and if it has a concrete and well defined body, it'll 'feel' like it's its own, however if it doesn't have a concrete body, it will feel like it doesn't have an ego. I've never done drugs so I don't know how it feels like, but when people describe ego death, I conjecture that the mechanisms in their brain that take the input of the body and wire it up with the consciousness are so confused that the consciousness can no longer make sense of them, thus it no longer feels like it has a body, it looses it's frame of reference, and thus its ego 'dies', and in this state of confusion 'having a body' either doesn't make sense or is negated, its negation being 'the body being everywhere', which would explain some people's feeling of becoming one with the universe or with all creatures at once or connecting with a higher plane etc etc. I'll try drugs one day.