Coding assistants
category: code [glöplog]
And with that said, I also totally recognize the pitfalls that Gargaj points out. It has to be used selectively to confer a boost.
Blueberry and me are talking about the "smart" autocomplete features. We are not talking about prompting and generating large portions of the code. I can totally understand the figure above for that case.
In case the autocomplete is too slow I just keep typing. but in many cases it is just quick enough especially for stupid stuff which just needs to be typed. obviously you will also have to micro review it but for stupid stuff it is absolutely great.
remove the first 3 bars and that is what we (or at least me) are talking about (also note how the bars would redistribute) :)
In case the autocomplete is too slow I just keep typing. but in many cases it is just quick enough especially for stupid stuff which just needs to be typed. obviously you will also have to micro review it but for stupid stuff it is absolutely great.
remove the first 3 bars and that is what we (or at least me) are talking about (also note how the bars would redistribute) :)
Quote:
obviously you will also have to micro review it
Quote:
remove the first 3 bars
*looks at the first bar*
mhmh...
Anyway, my main problem with AI is the CO2 emissions, water and electricity usage, and overall waste of resources. I think saving a few lines of manually typed code is not worth that. But no one seems to care, let's cook the planet...
Also, as per my last bullet point in the OP, there are situations where you are optimizing not for total time spent, but rather for e.g. screen time or keystrokes. Here coding assistants shine.
Quote:
It depends on how you use it. I generally find that spending less time on tedious typing allows me to spend a greater proportion of the coding time actually thinking about the code I write.
I actually did use AI for API-related queries myself - mostly because you really have no choice. Even if you ask search engine today it will most likely output you AI overview and the rest of the search results is crap. So chances are you use it without even knowing it.
Another example, I recently forgot how to setup FBO and was like "dear chatGPT".. but then I started to worry - that maybe Gargaj with his conservativism here is right. We were able to do it without any AI back in the days, so...and it wasn't such a big impediment.
And yeah, let's face it - we are as a community giving away too much for "the rich" who own those effing models. And they blatantly violate our rights (licenses), so it's not even close to be a fair game. This is not ok actually. I said it multiple times, but I was never big on open source and now I have zero open source policy for my projects - sorry, I see no other way.
PS not to mention, it's harder to pick a host for repository that does not steal your private code. And, if you use AI assistant in your coding IDE chances are you agreed to terms (with a small print) of using your private code for further improvements of the assistant.
Quote:
maybe Gargaj with his conservativism
"skepticism", if you'd be so kind.
This divide is too big to be conquered. Let’s vote, split the loot, bid farewell.
When painting an image, nearly every brushstroke involves an artistic decision. When writing code, there are a lot of chores between artistic decisions.
Automating just those chores is totally fine for me. Automating artistic decisions, not so much.
Automating just those chores is totally fine for me. Automating artistic decisions, not so much.
Quote:
When writing code, there are a lot of chores between artistic decisions.
If those things are chores to you, it means your code structure is bad. If you carefully design your code it's like a beautiful layers of brushes in all abstraction levels. It all takes time to learn, just like mastering the art of painting.
I would say there is even more decisions you have to take along the way than a painter. Even when implementing a simple dot product or orientation test, you have to take care of subtle numerical issues, etc...if you implement algorithm, it takes enormous skill to limit number of operations and computational complexity.
Now, I've seen people first-hand bastardizing the whole process with AI and claiming the "job done", but their code was a joke.
Quote:
Automating just those chores is totally fine for me.
There is really no clear cut line with AI assistance between speeding you up with simple API calls and solving complex algorithmic issue. And it's easy to get lazy here.
Again, if you want to make your overlords happy, sure, maybe you even like it, I'm not judging, but at least please be more aware of the consequences.
PS I'm totally ok if you develop LLM model yourself based on legit code and use it as AI assistant. I just seriously think we should not "fatten the pigs" more...they are already fat enough.
Quote:
it means your code structure is bad
<facepalm>
Quote:
Quote:it means your code structure is bad
<facepalm>
I should elaborate : if your arguments against AI-based IDE is telling your pairs about the supposedly bad quality of their code, well... not in my name :D
Quote:
If those things are chores to you, it means your code structure is bad. If you carefully design your code it's like a beautiful layers of brushes in all abstraction levels.
I want to see your beautiful code that gets along without anything that's a typical target of autocompletion such as overload method headers or import statements. The only case that comes to my mind where that might be the case is coding in a machine code monitor like in the early 80s :P
Ah ok, so you decided to attack the messenger.
And I want to see your beautiful art that is so original that cannot be created with simple diffusion model.
Look - that's not the point really. My (or anyone) code is of course far from perfect or 90% of it is not even original, but it's only thanks to code written by highly skilled coders those models can even work and "auto complete". I'm personally not willing to give up my rights to my creations, so somebody else can have unlimited right to exploit it, but seriously, if you are OK with it - go for it.
Quote:
I want to see your beautiful code
And I want to see your beautiful art that is so original that cannot be created with simple diffusion model.
Look - that's not the point really. My (or anyone) code is of course far from perfect or 90% of it is not even original, but it's only thanks to code written by highly skilled coders those models can even work and "auto complete". I'm personally not willing to give up my rights to my creations, so somebody else can have unlimited right to exploit it, but seriously, if you are OK with it - go for it.
Quote:
nearly every brushstroke involves an artistic decision.
Sure. For less noble actions I have servants. They inhale turpentine instead of me. After I determine the exact hues and needed quantities, they go and mix that. They prepare the canvas. They sketch the thing, they underpaint sh*t. Then I come and make art with every flick of my wrist.
No, srsly, I was under the impression that demoscene is, at least in a big part, exactly “the art of the boilerplate”. “Look mom, I improved and furtherly streamlined a 20 year old boilerplate code. Because I’m crazy. Because code is not chore, but art for me.” That kinda stuff. That this is the distinction between the demoscene and artsy fartsy modern gallery fodder.
The current “quality” of vibe code really doesn’t matter. The ability to determine if it’s AI made or not also doesn’t matter. I am of the opinion that there should be no will, no ambition to use AI in the demoscene. Not forbidden, but detested and shunned in the first place. It’s like using steroids in a community soccer match. It’s poisonous, just look at how it divides us. For one, I personally feel so strongly about this, that I sometimes really think we should split up in two over this issue this if there is no consensus. Unfortunately, I see many a long time friend and colleague through a different lens because of their indifference or their willingness to smear themselves with this crap.
To me, saying you can't use AI for coding demos is like drawing an arbitrary line in the sand and say "to qualify as demoscene works, you must only use tech from before this year". Why 2024? Why not, say, 1990? People code Amiga demos on modern laptops too, right? Nobody ever got banned from a compo for that, but it's the exact same thing: use modern tools, irrespective of the age of the platform.
I for one hope that maybe AI finally helps some people get over those first few burdens of getting into graphics coding. It really is very difficult to get started with. Maybe we'll see more creativity, rather than less? I'm not sure, but I sure hope so!
I for one hope that maybe AI finally helps some people get over those first few burdens of getting into graphics coding. It really is very difficult to get started with. Maybe we'll see more creativity, rather than less? I'm not sure, but I sure hope so!
Quote:
Maybe we'll see more creativity, rather than less? I'm not sure, but I sure hope so!
That’s a nice thought. I think we’ll see a flood of mediocre slop instead. But some will call it “democratization”. Some others “true art”. We’ve already seen the preview.
For the record, when I say chores, I mean just that - simple tasks that are usually done the same way such as creating an overload method header. Your vector math code would be an actual task, not a chore, something better left to humans who know what they're doing.
And no need to feed the cloud with your code, there's code completion that runs locally on your own machine.
And no need to feed the cloud with your code, there's code completion that runs locally on your own machine.
Quote:
get over those first few burdens of getting into graphics coding. It really is very difficult to get started with.
It's easier than it's ever been, honestly; there are an infinite amount of frameworks, toolkits, tutorials... if anything there's a bit of choice paralysis.
Quote:
For the record, when I say chores, I mean just that - simple tasks that are usually done the same way such as creating an overload method header.
Don't we already have classic AST-based helpers like VAX and Resharper for that?
Quote:
I think we’ll see a flood of mediocre slop instead.
...in code? How could you tell?
Choice paralysis is big yeah.
As much as I deeply hate most uses of AI, especially when it's fundamentally something very human or expressive, I do constantly use it for chores, reading docs and boilerplate. Also for studying/learning.
If it gets better, I'm absolutely not looking forward to vibecoded entries and that shit would stink.
Hell, I'm also not looking forward to cleaning/tidying robots, because that also is very human, expressive and meditational.
Reading the miniaudio docs isn't. Neither is remembering how to convert a wchar path to a char string in cpp. Nor is some obscure problem with QT that is very hard to google.
The art of coding doesn't lie here and I don't care where the evil robot scraped it from, but i would care once it starts writing anything which isn't basic garbage.
As much as I deeply hate most uses of AI, especially when it's fundamentally something very human or expressive, I do constantly use it for chores, reading docs and boilerplate. Also for studying/learning.
If it gets better, I'm absolutely not looking forward to vibecoded entries and that shit would stink.
Hell, I'm also not looking forward to cleaning/tidying robots, because that also is very human, expressive and meditational.
Reading the miniaudio docs isn't. Neither is remembering how to convert a wchar path to a char string in cpp. Nor is some obscure problem with QT that is very hard to google.
The art of coding doesn't lie here and I don't care where the evil robot scraped it from, but i would care once it starts writing anything which isn't basic garbage.
Quote:
I'm absolutely not looking forward to vibecoded entries and that shit would stink.
https://www.pouet.net/prod.php?which=94788
Quote:
Frankly, after the scene has grown up tolerating - if not actively embracing - remixes of everything from Rob Hubbard to Boris Vallejo to the Amen Break to Mezzoforte, I have to doubt the sincerity of anyone who suddenly decides that we now have to take an ultra-purist stance on LLM output that might theoretically contain traces of copyrighted code, just because it happens to support the current anti-AI narrative.
Remix has, you know, aknowledgment or a shoutout to the original author. Even if there’s no official acknowledgement, people hear/see and decide for themselves how to feel about it. Doing it with AI is extra sneaky. The AI “author” blatantly pretends it’s his own work. While it’s not. This is ethically deeply repulsive, but techbros are pushing it on us and saying that it’s not ethically dubious. I can’t believe people have such weak personal code of ethics that can so easily get corrupted by techbro Grima whispering in their ears, whispering about instant short term glory, while simultaneously setting the scene to crush them in near to mid future.
Plus, you said “grow up”. Exactly. You do stupid sh*t growing up. And then you grow up.
my personal perspective on this is that the stealing/copyright issues and the fundamentally non-human nature of ai (both pointed out in this thread already) are the biggest concerns related to the scene.
regarding the stealing, i don't think it's comparable with remix culture in any way, as what llms do for example is not like a human being picking a specific piece of art of another human being and using it in a specific way.
regarding the non-human nature of ai, i think that art is something made by humans and is made for humans. that's the very nature of it.
so, at the end of the day, all these questions of purity and just tools and libraries always culminate in the question where to draw the line. answers can be diverse and might shift with time as the impact of ai becomes more clear, but i believe the line has to be drawn.
regarding the stealing, i don't think it's comparable with remix culture in any way, as what llms do for example is not like a human being picking a specific piece of art of another human being and using it in a specific way.
regarding the non-human nature of ai, i think that art is something made by humans and is made for humans. that's the very nature of it.
so, at the end of the day, all these questions of purity and just tools and libraries always culminate in the question where to draw the line. answers can be diverse and might shift with time as the impact of ai becomes more clear, but i believe the line has to be drawn.
Quote:
the stealing/copyright issues
It's not just a simple old-fashioned copyright violation. It's a large scale violation.
Before AI, an average unaffiliated joe (by default) owned full rights to the code he wrote - no matter if the code was good or not (in anticipation to Kabuto's Stockholm syndrome), unless he willingly gave up his rights. But now, a relatively small group of people (backed by private investment firms) found a loophole to not only violate those rights, but effectively transfer them by reclaiming ownership of the models trained on violated materials.
So if you use their coding assistants locally - you are still violating copyrights of the respective creators (same applies to art/music and code really). And if you are using them online, chances are you are additionally giving away (unwillingly?) your intellectual property. Some people are totally fine with both for some reason that is hard for me to understand, but then again, some people like being humiliated and mistreated, so why not.
So, it turns out this "democratization" could well serve to furtherly lock away (human) public coding knowledge as more people start locking away and hiding their source code, unwilling to let the big tech exploit them. There was this big happy community that freely shared their secrets for mutual benefits and these motherfuckers come and attach their inhuman leach to that community source an deploy it for (future) profit. And, as @tomkh says: to that, some people say: "Yay! Rape me some more please." wtf
This is not AaronSwartzing, in fact it's the opposite. The "wild west / big democratization" phase is here only until their leach grows to size where it's too big to fail / shutdown. Like Google inc.
This is not AaronSwartzing, in fact it's the opposite. The "wild west / big democratization" phase is here only until their leach grows to size where it's too big to fail / shutdown. Like Google inc.