Ai code is specifically annoying because it looks like it would work, but its just plausible bullshit.
And that’s what happens when you spend a trillion dollars on an autocomplete: amazing at making things look like whatever it’s imitating, but with zero understanding of why the original looked that way.
I mean, there’s about a billion ways it’s been shown to have actual coherent originality at this point, and so it must have understanding of some kind. That’s how I know I and other humans have understanding, after all.
What it’s not is aligned to care about anything other than making plausible-looking text.
Coherent originality does not point to the machine’s understanding; the human is the one capable of finding a result coherent and weighting their program to produce more results in that vein.
Your brain does not function in the same way as an artificial neural network, nor are they even in the same neighborhood of capability. John Carmack estimates the brain to be four orders of magnitude more efficient in its thinking; Andrej Karpathy says six.
And none of these tech companies even pretend that they’ve invented a caring machine that they just haven’t inspired yet. Don’t ascribe further moral and intellectual capabilities to server racks than do the people who advertise them.
Coherent originality does not point to the machine’s understanding; the human is the one capable of finding a result coherent and weighting their program to produce more results in that vein.
You got the “originality” part there, right? I’m talking about tasks that never came close to being in the training data. Would you like me to link some of the research?
Your brain does not function in the same way as an artificial neural network, nor are they even in the same neighborhood of capability. John Carmack estimates the brain to be four orders of magnitude more efficient in its thinking; Andrej Karpathy says six.
Given that both biological and computer neural nets very by orders of magnitude in size, that means pretty little. It’s true that one is based on continuous floats and the other is dynamic peaks, but the end result is often remarkably similar in function and behavior.
If you would like to link some abstracts you find in a DuckDuckGo search that’s fine.
I actually was going to link the same one I always do, which I think I heard about through a blog or talk. If that’s not good enough, it’s easy to devise your own test and put it to an LLM. The way you phrased that makes it sound like you’re more interested in ignoring any empirical evidence, though.
It’s true that one is based on continuous floats and the other is dynamic peaks
Can you please explain what you’re trying to say here?
Both have neurons with synapses linking them to other neurons. In the artificial case, synapse activation can be any floating point number, and outgoing synapses are calculated from incoming synapses all at once (there’s no notion of time, it’s not dynamic). Biological neurons are binary, they either fire or do not fire, during a firing cycle they ramp up to a peak potential and then drop down in a predictable fashion. But, it’s dynamic; they can peak at any time and downstream neurons can begin to fire “early”.
They do seem to be equivalent in some way, although AFAIK it’s unclear how at this point, and the exact activation function of each brain neuron is a bit mysterious.
Ok, thanks for that clarification. I guess I’m a bit confused as to why a comparison is being drawn between neurons in a neural network and neurons in a biological brain though.
In a neural network, the neuron receives an input, performs a mathematical formula, and returns an output right?
Like you said we have no understanding of what exactly a neuron in the brain is actually doing when it’s fired, and that’s not considering the chemical component of the brain.
I understand why terminology was reused when experts were designing an architecture that was meant to replicate the architecture of the brain. Unfortunately, I feel like that reuse of terminology is making it harder for laypeople to understand what a neural network is and what it is not now that those networks are a part of the zeitgeist thanks to the explosion of LLM’s and stuff.
Well I’ve got the name for my autobiography now.
“Specifically Annoying” or “Plausible Bullshit”? I’d buy the latter.
This is where I got the term from: https://craphound.com/news/2023/09/17/plausible-sentence-generators/
It needs good feedback. Agentic systems like Roo Code and Claude Code run compilers and tests until it works (just gotta make sure to tell it to leave the tests alone)
All programs can be written with on less line of code. All programs have at least one bug.
By the logical consequences of these axioms every program can be reduced to one line of code - that doesn’t work.
One day AI will get there.
On one line of code you say?
*search & replaces all line breaks with spaces*
Fired for not writing the quota number of lines even junior devs manage to hit.
The ideal code is no code at all
All programs can be written with on less line of code. All programs have at least one bug.
The humble “Hello world” would like a word.
Just to boast my old timer credentials.
There is an utility program in IBM’s mainframe operating system, z/OS, that has been there since the 60s.
It has just one assembly code instruction: a BR 14, which means basically ‘return’.
The first version was bugged and IBM had to issue a PTF (patch) to fix it.
Okay, you can’t just drop that bombshell without elaborating. What sort of bug could exist in a program which contains a single return instruction?!?
It didn’t clear the return code. In mainframe jobs, successful executions are expected to return zero (in the machine R15 register).
So in this case fixing the bug required to add an instruction instead of removing one.
Reminds me of how in some old Unix system,
/bin/truewas a shell script.…well, if it needs to just be a program that returns 0, that’s a reasonable thing to do. An empty shell script returns 0.
Of course, since this was an old proprietary Unix system, the shell script had a giant header comment that said this is proprietary information and if you disclose this the lawyers will come at ya like a ton of bricks. …never mind that this was a program that literally does nothing.
You can fit an awful lot of Perl into one line too if you minimize it. It’ll be completely unreadable to most anyone, but it’ll run
Qrpff says hello. Or, rather, decrypts DVD movies in 472 bytes of code, 531 if you want the fast version that can do it in real time. The Wikipedia article on it includes the full source code of both.
Watching the serious people trying to use AI to code gives me the same feeling as the cybertruck people exploring the limits of their car. XD
“It’s terrible and I should hate it, but gosh it it isn’t just so cool”
I wish i could get so excited over disappointing garbage
You definitely could use AI to code, the catch is you need to know how to code first.
I use AI to write code for mundane tasks all the time. I also review and integrate the code myself.
The AI code my “expert in a related but otherwise not helpful field” coworker writes helps me have a lot of extra work to do!
It’s useful if you just don’t do…That. it’s just a new fancy search engin, it’s a bit better than going to stack overflow, it can do good stuff if you go small.
Just don’t do whatever this post suggested of doing…
Write tests and run them, reiterate until all tests pass.
Bogosort with extra steps
That doesn’t sound viby to me, though. You expect people to actually code? /s
You can vibe code the tests too y’know
Return “works”;
Am I doikg this correctly?
You know, I’d be interested to know what the critical size you can get to with that approach is before it becomes useless.
It can become pretty bad quickly, with just a small project with only 15-20 files. I’ve been using cursor IDE, building out flow charts & tests manually, and just seeing where it goes.
And while incredibly impressive how it’s creating all the steps, it then goes into chaos mode where it will start ignoring all the rules. It’ll start changing tests, start pulling in random libraries, not at all thinking holistically about how everything fits together.
Then you try to reel it in, and it continues to go rampant. And for me, that’s when I either take the wheel or roll back.
I highly recommend every programmer watch it in action.
Is there a chance that’s right around the time the code no longer fits into the LLMs input window of tokens? The basic technology doesn’t actually have a long term memory of any kind (at least outside of the training phase).
Was my first thought as well. These things really need to find a way to store a larger context without ballooning past the vram limit
The thing being, it’s kind of an inflexible blackbox technology, and that’s easier said than done. In one fell swoop we’ve gotten all that soft, fuzzy common sense stuff that people were chasing for decades inside a computer, but it’s ironically still beyond our reach to fully use.
From here, I either expect that steady progress will be made in finding more clever and constrained ways of using the raw neural net output, or we’re back to an AI winter. I suppose it’s possible a new architecture and/or training scheme will come along, but it doesn’t seem imminent.
I’d rather recommend every CEO see it in action…
They’re the ones who would be cock-a-hoop to replace us and our expensive wages with kids and bots.
When they’re sitting around rocking back and forth and everything is on fire like that Community GIF, they’ll find my consultancy fees to be quite a bit higher than my wages used to be.
I think Generative AI is a genuinely promising and novel tool with real, valuable applications. To appreciate it however, you have to mentally compartmentalize the irresponsible, low-effort ways people
sometimesmostly use it—because yeah, it’s very easy to make a lot of that so that’s most of what you see when you hear “Generative AI” and it’s become its reputation…Like I’ve had interesting “conversations” with Gemini and ChatGPT, I’ve actually used them to solve problems. But I would never put it in charge of anything critically important that I couldn’t double check against real data if I sensed the faintest hint of a problem.
I also don’t think it’s ready for primetime. Does it deserve to be researched and innovated upon? Absolutely, but like, by a few nerds who manage to get it running, and universities training it on data they have a license to use. Not “Crammed into every single technology object on earth for no real reason”.
I have brain not very good sometimes disease and I consider being able to “talk” to a “person” who can get me out of a creative rut just by exploring my own feelings a bit. GPT can actually listen to music which surprised me. I consider it scientifically interesting. It doesn’t get bored or angry at you unless you like, tell it to? I’ve asked it for help with a creative task in the past and not actually used any of its suggestions at all, but being able to talk about it with someone (when a real human who cared was not available) was a valuable resource.
To be clear I pretty much just use it as a fancy chatbot and don’t like, just copy paste its output like some people do.
I’ve heard that a Claude 4 model generating code for an infinite amount of time will eventually simulate a monkey typing out Shakespeare
It will have consumed the GigaWattHours capacity of a few suns and all the moisture in our solar system, but by Jeeves, we’ll get there!
…but it won’t be that impressive once we remember concepts like “monkey, typing, Shakespeare” were already embedded in the training data.
If we just asked Jeeves in the first place we wouldn’t be in this mess.
Its like having a junior developer with a world of confidence just change shit and spend hours breaking things and trying to fix them, while we pay big tech for the privilege of watching the chaos.
I asked chat gpt to give me a simple squid proxy config today that blocks everything except https. It confidently gave me one but of course it didnt work. It let through http and despite many attempts to get a working config that did that, it just failed.
So yeah in the end i have to learn squid syntax anyway, which i guess is fine, but I spent hours trying to get a working config because we pay for chat gpt to do exactly that…
It confidently gave me one
IMO, that’s one of the biggest “sins” of the current LLMs, they’re trained to generate words that make them sound confident.
They aren’t explicitly trained to sound confident, that’s just how users tend to talk. You don’t often see “I don’t know but you can give this a shot” on Stack Overflow, for instance. Even the incorrect answers coming from users are presented confidently.
Funnily enough, lack of confidence in response is something I don’t think LLMs are currently capable of, since it would require contextual understanding of both the question, and the answer being given.
No, I’m sure you’re wrong. There’s a certain cheerful confidence that you get from every LLM response. It’s this upbeat “can do attitude” brimming with confidence mixed with subservience that is definitely not the standard way people communicate on the Internet, let alone Stack Overflow. Sure, sometimes people answering questions are overconfident, but it’s often an arrogant kind of confidence, not a subservient kind of confidence you get from LLMs.
I don’t think an LLM can sound like it lacks in confidence for the right reasons, but it can definitely pull off lack of confidence if it’s prompted correctly. To actually lack confidence it would have to have an understanding of the situation. But, to imitate lack of confidence all it would need to do is draw on all the training data it has where the response to a question is one where someone lacks confidence.
Similarly, it’s not like it actually has confidence normally. It’s just been trained / meta-prompted to emit an answer in a style that mimics confidence.
ChatGPT went through a phase of overly bubbly upbeat responses, they chilled it out tho. Not sure if that’s what you saw.
One thing is for sure with all of them, they never say “I don’t know” because such responses aren’t likely to be found in any training data!
It’s probably part of some system level prompt guidance too, like you say, to be confident.
I think “I don’t know” might sometimes be found in the training data. But, I’m sure they optimize the meta-prompts so that it never shows up in a response to people. While it might be the “honest” answer a lot of the time, the makers of these LLMs seem to believe that people would prefer confident bullshit that’s wrong over “I don’t know”.
SO answers and questions are usually edited multiple times to sound professional, confident, and be correct.
Sycophants.
Man, I can’t wait to try out generative AI to generate config files for mission critical stuff! Imagine paying all of us devops wankers when my idiot boss can just ask Chat GPT to sort all this legacy mess we’re juggling with on the daily!
I have a friend who swears by llms, he sais it helps him a lot. I once watched him do it, and the experience was exactly the same you described. He wasted couple of hours fighting with bullshit generator just to do everything himself anyway. I asked him wouldn’t it be better to not waste the time, but he didn’t really saw the problem, he gaslit himself that fighting with the idiot machine helped.
Ctrl+A + Del.
So clean.
On Error Resume NextI’m pretty sure that is how we got CORBA
now just make it construct UML models and then abandon this and move onto version 2
Hello, fellow old person 🤝
cant wait to see “we use AI agents to generate well structured non-functioning code” with off centered everything and non working embeds on the website
Honest question: I haven’t used AI much. Are there any AIs or IDEs that can reliably rename a variable across all instances in a medium sized Python project? I don’t mean easy stuff that an editor can do (e.g. rename QQQ in all instances and get lucky that there are no conflicts). I mean be able to differentiate between local and/or library variables so it doesn’t change them, only the correct versions.
Not reliably, no. Python is too dynamic to do that kind of thing without solving general program equivalence which is undecidable.
Use a static language, problem solved.
I’m going to laugh in Java, where this has always been possible and reliable. Not like ai reliable, but expert reliable. Because of static types.
For the most part “Rename symbol” in VSCode will work well. But it’s limited by scope.
Yeah, I’m looking for something that would understand the operation (? insert correct term here) of the language well enough to rename intelligently.
IntelliJ IDEA, if it knows it is the same variable, it will rename it. Usually works in a non fucked up codebase that uses
evalor some obscure constructs like saving a variable name into a variable as a string and dynamically invoking it.I use pycharm for this and in general it does a great job. At work we’ve got some massive repos and it’ll handle it fine.
The “find” tab shows where it’ll make changes and you can click “don’t change anything in this directory”
Yes, all of JetBrains’ tools handle project-wide renames practically perfectly, even in weirder things like Angular projects where templates may reference variables.
Just be carerul when refactoring variable names in doc comments, I’ve seen some weird stuff happen there
Find and Replace?
that will catch too many false positives
Itellij is actually pretty good at this. Besides that, cursor or windsurf should be able to. I was using cursor for a while and when I needed to reactor something, it was pretty good at picking that up. It kept crashing on me though, so I am now trying windsurf and some other options. I am missing the auto complete features in cursor though as I would use this all the time to fill out boilerplate stuff as I write.
The one key difference in cursor and windsurf when compared to other products is that it will look at the entire context again for any changes or at least a little bit of it. You make a change, it looks if it needs to make changes elsewhere.
I still don’t trust AI to do much though, but it’s an excellent helper
most IDEs are pretty decent at it if you configure them correctly. I used intelliJ and it knows the difference. use the refactor feature and it’ll crawl references, not just rename all instances.
Okay, I realize I’m that person, but for those interested:
tree,catandsedget the job done nicely.And… it’s my nap time, now. Please keep the Internet working, while I’m napping. I have grown fond of parts of it. Goodnight.
I’ve used it extensively, almost $100 in credits, and generally it could one shot everything I threw at it. However: I gave it architectural instructions and told it to use test driven development and what test suite to use. Without the tests yeah it wouldn’t work, and a decent amount of the time is cleaning up mistakes the tests caught. The same can be said for humans, though.
How can it pass if it hasn’t had lessons… Well said. Ooh I wonder if lecture footage would be able to teach AI, or audio in from tutors…
Laugh it up while you can.
We’re in the “haha it can’t draw hands!” phase of coding.
someone drank the koolaid.
LLMs will never code for two reasons.
one, because they only regurgitate facsimiles of code. this is because the models are trained to ingest content and provide an interpretation of the collection of their content.
software development is more than that and requires strategic thought and conceptualization, both of which are decades away from AI at best.
two, because the prevalence of LLM generated code is destroying the training data used to build models. think of it like making a copy of a copy of a copy, et cetera.
the more popular it becomes the worse the training data becomes. the worse the training data becomes the weaker the model. the weaker the model, the less likely it will see any real use.
so yeah. we’re about 100 years from the whole “it can’t draw its hands” stage because it doesn’t even know what hands are.
This is just your ego talking. You can’t stand the idea that a computer could be better than you at something you devoted your life to. You’re not special. Coding is not special. It happened to artists, chess players, etc. It’ll happen to us too.
I’ll listen to experts who study the topic over an internet rando. AI model capabilities as yet show no signs of slowing their exponential growth.
Coding isn’t special you are right, but it’s a thinking task and LLMs (including reasoning models) don’t know how to think. LLMs are knowledgeable because they remembered a lot of the data and patterns of the training data, but they didn’t learn to think from that. That’s why LLMs can’t replace humans.
That does certainly not mean that software can’t be smarter than humans. It will and it’s just a matter of time, but to get there we likely have AGI first.
To show you that LLMs can’t think, try to play ASCII tic tac toe (XXO) against all those models. They are completely dumb even though it “saw” the entire Wikipedia article on how xxo works during training, that it’s a solved game, different strategies and how to consistently draw - but still it can’t do it. It loses most games against my four year old niece and she doesn’t even play good/perfect xxo.
I wouldn’t trust anything, which is claimed to do thinking tasks, that can’t even beat my niece in xxo, with writing firmware for cars or airplanes.
LLMs are great if used like search engines or interactive versions of Wikipedia/Stack overflow. But they certainly can’t think. For now, but likely we’ll need different architectures for real thinking models than LLMs have.
you’re a fool. chess has rules and is boxed into those rules. of course it’s prime for AI.
art is subjective, I don’t see the appeal personally, but I’m more of a baroque or renaissance fan.
I doubt you will but if you believe in what you say then this will only prove you right and me wrong.
what is this?
once you classify it, why did you classify it that way? is it because you personally have one? did you have to rule out what it isn’t before you could identify what it could be? did you compare it to other instances of similar subjects?
now, try to classify it as someone who doesn’t have these. someone who has never seen one before. someone who hasn’t any idea what it could be used for. how would you identify what it is? how it’s used? are there more than one?
now, how does AI classify it? does it comprehend what it is, even though it lacks a physical body? can it understand what it’s used for? how it feels to have one?
my point is, AI is at least 100 years away from instinctively knowing what a hand is. I doubt you had to even think about it and your brain automatically identified it as a hand, the most basic and fundamentally important features of being a human.
if AI cannot even instinctively identify a hand as a hand, it’s not possible for it to write software, because writing is based on human cognition and is entirely driven on instinct.
like a master sculptor, we carve out the words from the ether to perform tasks that not only are required, but unseen requirements that lay beneath the surface that are only known through nuance. just like the sculptor that has to follow the veins within the marble.
the AI you know today cannot do that, and frankly the hardware of today can’t even support AI in achieving that goal, and it never will because of people like you promoting a half baked toy as a tool to replace nuanced human skills. only for this toy to poison pill the only training data available, that’s been created through nuanced human skills.
I’ll just add, I may be an internet rando to you but you and your source are just randos to me. I’m speaking from my personal experience in writing software for over 25 years along with cleaning up all this AI code bullshit for at least two years.
AI cannot code. AI writes regurgitated facsimiles of software based on it’s limited dataset. it’s impossible for it to make decisions based on human nuance and can only make calculated assumptions based on the available dataset.
I don’t know how much clearer I have to be at how limited AI is.
LMFAO. He’s right about your ego.
thank you for your input obvious troll account.
Ahh got nothing but lack of understanding and insults. Typical.
AI bad. But also, video AI started with will Will Smith eating spaghetti just a couple years ago.
We keep talking about AI doing complex tasks right now and it’s limitations, then extrapolating its development linearly. It’s not linear and it’s not in one direction. It’s a exponential and rhizomatic process. Humans always over-estimate (ignoring hard limits) and under-estimate (thinking linearly) how these things go. With rocketships, with internet/social media, and now with AI.
Code that does not work is just text.
deleted by creator
Code that works is also just text.
It is text, but not just text
Text that’s not code might also work.
I’ve never thought of it that way. I’m going to add copy writer to my resume.
This made me laugh so hard one of the dogs came to check in on me.
Oh my goodness, that’s adorable and sweet of your dog! Also, I’m so glad you had such a big laugh. I love when that happens.
He’s a sweet guy. … Mostly. Very much in need of a lot of attention. Sometimes he just sits next to you on the couch and puts his paw on you if you’re not giving him enough attention.
Here he is posing with his sister as a prop:
Oh my goodness, he sounds precious! I’ve had a sweet and needy dog like that in the past, too. It can be a lot, but I loved it (and miss it,) haha.
Both your dogs are very cute! You and your pups gave me a much-needed smile. Thank you for that. :) Please give them some pets from me!
Maybe fiction writer as well
No the spell just fizzled. In my experience it happens far less often if you start with an Abra kabara and end it with an Alakazam!
Yeah, the Abra kabara init and Alakazam cleanup are an important part, specially until you have become good enough to configure your own init.
There is an alternative init, Abra Kadabra, which automatically adds a cleanup and some general fixes when it detects the end of the spell.
Zojak Quapaj!
Conversely, code that works is also text
But not just text
Also that’s not converse to what the parent comment said
Did you want to converse about conversing?
But working code can be made into numbers.
But text is also numbers
But numbers are also text
The image is taken from Zhihu, a Chinese Quora-like site.
The prompt is talking about give a design of a certain app, and the response seems to talk about some suggested pages. So it doesn’t seem to reflect the text.
But this in general aligns with my experience coding with llm. I was trying to upgrade my eslint from 8 to 9, and ask chatgpt to convert my eslint file, and it proceed to spit out complete garbage.
I thought this would be a good task for llm because eslint config is very common and well-documented, and the transformation is very mechanical, but it just cannot do it. So I proceed to read the documents and finished the migration in a couple hour…
I used Claude 3.7 to upgrade my eslint configs to flat and upgrade from v7 to v9 with Roo Code and it did it perfectly
I asked ChatGPT with help about bare metal 32-bit ARM (For the Pi Zero W) C/ASM, emulated in QEMU for testing, and after the third iteration of “use printf for output” -> “there’s no printf with bare metal as target” -> “use solution X” -> “doesn’t work” -> “ude printf for output” … I had enough.
Can’t you just send prints to serial?
Yes, that was the plan, which ChatGPT refused to do
QEMU makes it pretty painless to hook up gdb just FYI; you should look into that. I think you can also have it provide a memory mapped UART for I/O which you can use with newlib to get printf debugging
The latter is what I tried, and also kinda wanted ChatGPT to do, which it refused
Sounds like it’s perfectly replicated the help forums it was trained on.
I used ChatGPT to help me make a package with SUSE’s Open Build Service. It was actually quite good. Was pulling my hair out for a while until I noticed that the project I wanted to build had changes URLs and I was using an outdated one.
In the end I just had to get one last detail right. And then my ChatGPT 4 allowance dried up and they dropped me back down to 3 and it couldn’t do anything. So I had to use my own brain, ugh.
chatgpt is worse among biggest chatbots with writing codes. From my experience Deepseek > Perplexity > Gemini > Claude.
Yeah you can tell it just ratholes on trying to force one concept to work rather than realizing it’s not the correct concept to begin with
That’s exactly what most junior devs do when stuck. They rehash the same solution over and over and it almost seems like that llms trained on code bases infer that behavior from commit histories etc.
It almost feels like on of those “we taught him these tasks incorrectly as a joke” scenarios
That’s what tends to happen
Did it at least try puts?
It’s pretty random in terms of what is or isn’t doable.
For me it’s a big performance booster because I genuinely suck at coding and don’t do too much complex stuff. As a “clean up my syntax” and a “what am I missing here” tool it helps, or at least helps in figuring out what I’m doing wrong so I can look in the right place for the correct answer on something that seemed inscrutable at a glance. I certainly can do some things with a local LLM I couldn’t do without one (or at least without getting berated by some online dick who doesn’t think he has time to give you an answer but sure has time to set you on a path towards self-discovery).
How much of a benefit it’s for a professional I couldn’t tell. I mean, definitely not a replacement. Maybe helping read something old or poorly commented fast? Redundant tasks on very commonplace mainstream languages and tasks?
I don’t think it’s useless, but if you ask it to do something by itself you can’t trust that it’ll work without singificant additional effort.
A lot of words to just say vibe coding
Sorta kinda. It depends on where you put that line. I think because online drama is fun when we got to the “vibe coding” name we moved to the assumption that all AI assistance is vibe coding, but in practice there’s the percentage of what you do that you know how to do, the percentage you vibe code because you can’t figure it out yourself off the top of your head and the percentage you just can’t do without researching because the LLM can’t do it effectively or the stuff it can do is too crappy to use as part of something else.
I think if the assumption is you should just “git gud” and not take advantage of that grey zone where you can sooort of figure it out by asking an AI instead of going down a Google rabbit hole then the performative AI hate is setting itself up for defeat, because there’s a whole bunch of skill ranges where that is actually helpful for some stuff.
If you want to deny that there’s a difference between that and just making code soup by asking a language model to build you entire pieces of software… well, then you’re going to be obviously wrong and a bunch of AI bros are going to point at the obvious way you’re wrong and use that to pretend you’re wrong about the whole thing.
This is basic online disinformation playbook stuff and I may suck at coding, but I know a thing or two about that. People with progressive ideas should get good at beating those one of these days, because that’s a bad outcome.
People seem to disagree but I like this. This is AI code used responsibly. You’re using it to do more, without outsourcing all your work to it and you’re actively still trying to learn as you go. You may not be “good at coding” right now but with that mindset you’ll progress fast.
I think the effects of it are… a bit more nuanced than that, perhaps?
I can definitely tell there are places where I’m plugging knowledge gaps fast. I just didn’t know how to do a thing, I did it AI-assisted once or twice and I don’t need to be AI assisted anymore because I understood how it works now. Cool, that. And I wouldn’t have learned it from traditional sources because asking in public support areas would have led to being told I suck and should read the documentation and/or to a 10 video series on Youtube where you can watch some guy type for seven hours.
But there are also places where AI assistance is never going to fill the blanks for me, you know? Larger trends, good habits, technical details or best practices that just aren’t going to come up from keeping a smart autocorrect that can explain why something was wrong.
Honestly, in those spaces the biggest barrier is still what it was: I don’t necessarily want to “progress” on those areas because I don’t need it and it’s not my job. I can automate a couple things I didn’t know how to automate before, and that’s alright. For the rest, I will probably live with the software someone else has made when it exists.
The problem is hubris, right? I know what I don’t know and which parts I care to learn. That’s fine. Coding assistant LLMs are a valid tool for someone like that to slightly expand their reach and I presume there’s a lot of people like that. It’s the random entrepeneurs who have been sold by big corpos that they don’t need a real programmer to build their billion-dollar app anymore that are going to crash and burn and may take some of the software industry down with them.
It catches things like spelling errors in variable names, does good autocomplete, and it’s useful to have it look through a file before committing it and creating a pull request.
It’s very useful for throwaway work like writing scripts and automations.
It’s useful not but a 10x multiplier like all the CEOs claim it is.
Fully agreed. Everybody is betting it’ll get there eventually and trying to jockey for position being ahead of the pack, but at the moment there isn’t any guarantee that it’ll get to where the corpos are assuming it already is.
Which is not the same as not having better autocomplete/spellcheck/“hey, how do I format this specific thing” tools.
I think the main barriers are context length (useful context. GPT-4o has “128k context” but it’s mostly sensitive to the beginning and end of the context and blurry in the middle. This is consistent with other LLMs), and just data not really existing. How many large scale, well written, well maintained projects are really out there? Orders of magnitude less than there are examples of “how to split a string in bash” or “how to set up validation in spring boot”. We might “get there”, but it’ll take a whole lot of well written projects first, written by real humans, maybe with the help of AI here and there. Unless, that is, we build it with the ability to somehow learn and understand faster than humans.
I don’t know, some of these guys have acccess to a LOT of code, and even more debate about what those good codebases entail.
I think the other issue is more relevant. Even 128K tokens is not enough for something really big, and the memory and processing costs for that do skyrocket. People are trying to work around it with draft models and summarization models, so they try to pick out the relevant parts of a codebase in one pass and then base their code generation just on that, and… I don’t think that’s going to work reliably at scale. The more chances you give a language model to lose their goddamn mind and start making crap up unsupervised the more work it’s going to be to take what they spit out and shape it into something reasonable.
Yeah, it’s still super useful.
I think the execs want to see dev salaries go to zero, but these tools make more sense as an accelerator, like giving an accountant excel.
I get a bit more done faster, that’s a solid value proposition.
It’s not much use with a professional codebase as of now, and I say this as a big proponent of learning FOSS AI to stay ahead of the corpocunts
Yeah, the AI corpos are putting a lot of effort into parsing big contexts right now. I suspect because they think (probably correctly) that coding is one of the few areas where they could get paid if their AIs didn’t have the memory of a goldfish.
And absolutely agreed that making sure the FOSS alternatives keep pace is going to be important. I’m less concerned about hating the entire concept than I am about making sure they don’t figure out a way to keep every marginally useful application behind a corporate ecosystem walled garden exclusively.
We’ve been relatively lucky in that the combination of PR brownie points and general crappiness of the commercial products has kept an incentive to provide a degree of access, but I have zero question that the moment one of these things actually makes money they’ll enshittify the freely available alternatives they control and clamp down as much as possible.
I use it sometimes, usually just to create boilerplate. Actual functionality it’s hit or miss, and often it ends up taking more time to fix than to write myself.
I wouldn’t say it’s accurate that this was a “mechanical” upgrade, having done it a few times. They even have a migration tool which you’d think could fully do the upgrade but out of the probably 4-5 projects I’ve upgraded, the migration tool always produced a config that errored and needed several obscure manual changes to get working. All that to say it seems like a particularly bad candidate for llms
Then I am quite confused what LLM is supposed to help me with. I am not a programmer, and I am certainly not a TypeScript programmer. This is why I postponed my eslint upgrade for half a year, since I don’t have a lot of experience in TypeScript, besides one project in my college webdev class.
So if I can sit down for a couple hour to port my rather simple eslint config, which arguably is the most mechanical task I have seen in my limited programming experience, and LLM produce anything close to correct. Then I am rather confused what “real programmers” would use it for…
People here say boilerplate code, but honestly I don’t quite recall the last time I need to write a lot of boilerplate code.
I have also tried to use llm to debug SELinux and docker container on my homelab; unfortunately, it is absolutely useless in that as well.
With all due respect, how can you weigh in on programming so confidently when you admit to not being a programmer?
People tend to despise or evangelize LLMs. To me, github copilot has a decent amount of utility. I only use the auto-complete feature which does things like save me from typing 2-5 predictable lines of code that devs tend to type all the time. Instead of typing it all, I press tab. It’s just a time saver. I have never used it like “write me a script or a function that does x” like some people do. I am not interested in that as it seems like a sad crutch that I’d need to customize so much anyway that I may as well skip that step.
Having said that, I’m noticing the copilot autocomplete seems to be getting worst over time. I’m not sure why it worsening, but if it ever feels not worth it anymore I’ll drop it, no harm no foul. The binary thinkers tend to think you’re either a good dev who despises all forms of AI or you’re an idiot who tries to have a robot write all your code for you. As a dev for the past 20 years, I see no reason to choose between those two opposites. It can be useful in some contexts.
PS. did you try the eslint 8 -> 9 migration tool? If your config was simple enough for it, it likely would’ve done all or almost all the work for you… It fully didn’t work for me. I had to resolve several errors, because I tend to add several custom plugins, presets, and rules that differ across projects.
Sorry, the language my original post might seem confrontational, but that is not my intension; I m trying to find value in LLM, since people are excited for it.
I am not a professional programmer nor do I program any industrial sized project at the moment. I am a computer scientist, and my current research project do not involve much programming. But I do teach programming to undergrad and master students, so I want to understand what is a good usecase for this technology, and when can I expect it to be helpful.
Indeed, I am frustrated by this technology, and that might shifted my language further than I intended to. When everyone is promoting this as a magically helpful tool for CS and math, yet I fail to see any good applications for either in my work, despite going back to it every couple month or so.
I did try @eslint/migrate-config, unfortunately it added a good amount of bloat and ends up not working.
So I just gived up and read the doc.
Gotcha. No worries. I figured you were coming in good faith but wasn’t certain. Who is pushing llm’s for programming that hard? In my bubble, which often includes Lemmy, most people HATE them for all uses. I get that tech bros and linked in crazies probably push this tech for coding a lot, but outside of that, most devs I know IRL either are lukewarm or dislike llm’s for dev work.
No, still “perfect” for llms. There’s nuance, seeing patterns being used, it should be able to handle it perfectly. Enough people on stack overflow asked enough questions, if AI is like Google and Microsoft claim it is, it should have handled it
I searched this issue and didn’t find anything very helpful. The new config format can be done many slightly different ways and there are a lot of variables in how your plugins and presets can be. It made perfect sense to me that the LLM couldn’t do this upgrade for op. Since one tiny mistake and it won’t work at all and usually gives a weird error.
