Absolutely. It’s not fun to see what those AI vomit out these days.
Fully agreed
I’ve been using LLMs since I had to make them myself. And I started testing ChatGPT’s ability to code the week it came out. In the past month, I’ve seen so many articles about Claude.ai, that I decided to test it out for the second time. (The first time was when the term “vibe coding” had just been coined and people were making games with Claude.)
Claude is actually much better at some things than I expected. It still hallucinates, and it doesn’t always give the best answer IMO, but I have found it useful for small tasks like writing batch files. (I had to rename a few thousand audio files to snake_case and after some prompting, it wrote a perfectly capable script.)
My biggest problem is it goes above any beyond. I’ll be trying to remember the name of a function and so I’ll explain the function and tell it what I’m looking for, and it’ll pop back like a dozen possibilities with code examples, when all I wanted was the class name so I could go look it up.
Basically though, I use it like a search engine. Most recently I was working on importing KayKit models with a scrip, and I wanted the Barbarian and Barbarian_Large models to share the same material. So I checked every single KayKit model and found that when KayKit models use the same material, the always have an underscore to differentiate the types. Otherwise they are PascalCase. So I googled “godot 4 remove an underscore and any characters after it at the end of a string” and got bupkiss. The results weren’t super helpful, and the Google AI summary gave me like a 20-line solution. I knew this could be done as a one-liner.
Then I asked Claude the same thing. It output this:
Then it gave me 63 additional lines of example code.
What a waste of power and energy!
Here’s how I used it:
var material_path: String = GameConstants.MATERIALS_PATH + "kay_kit_models/" + str(owner_scene.name).rsplit("_", true, 1)[0].to_snake_case() + ".material"
Of course, if I’d given it context, I doubt I would have gotten a single line of code out of it.
All this is to say, that if you scope it down enough, I think LLMs can be useful. Do I let it write my code? Hell no! But if I was sharing it, would I tag my code as created or assisted by AI? No. Should I? Well that’s where we get into the ethical guidelines of this. Which is why I bring it up. Should my code be tagged as AI-assisted? And if so, is it unethical if I forgot that I asked Claude instead of Google this question?
I’d also like to add that I find it incredibly strange / weird that every time heavy AI users are asked to just… simply mark their work as ai assisted, they get incredibly upset, not understanding why they need to do that.
We’re not asking you to never post anything made with AI. We simply ask you to mark it. Why get upset that you need to do that?
Sounds very similar to this situation:
This is just a guess but I think they want to feel like real programmers and us making them mark it singles them out as not being the same as normal programmers
Yeah I see it’s use for minor stuff. But programming has always been about solving what is basically a puzzle. And so many people just let AI do that now so for me the benefits just aren’t enough to justify it. I just want it gone. It has uses, but it has way too many cons for me to accept it.
I would even love to see a tag stating which LLM was used for coding. I can see big differences in quality with different models.
Having used LLMs since ChatGPT 3.5, I can see really big improvements in Opus 4.6.
Recently I’ve been even mostly doing agentic coding for my flight simulator projecet. Which means the agent does all coding not only suggesting single lines of code.
When it comes to code quality I’ve only briefly looked over what the LLM produces. It mostly looks quite clean to me. As a seasoned programmer I’ve seen so much sloppy code made by humans who then still glorify them selfs, that I can live with AI beeing sloppy inbetween.
But I’m currious how far agentic coding will get me before I have to code myself again. The speed in development is uncanny.
I don’t think a tag is the best for that. The tag can just say AI or maybe if you want more detail I’d it was generative AI or an llm. But saying which model would be a LOT of tags so the person can just specify in the description of whatever asset/plugin it is which model was used.
Which (or what is the name of) LLM is the one shown in the short blurb at the top of a google search?
Never mind, I think Google search answered this:
The Gemini family of models powers AI-enhanced features in Google Search.
*Gemini AI used in this post.
What if we flipped the script and only labeled things ‘100% Manually Coded’ instead? If we assume AI is now a standard part of the modern developer’s toolkit, then AI-assistance is the new baseline. It’s actually more logical to label the exception (pure human effort) rather than the norm.
So make everyone who posts here do more work to say that their code isn’t LLM-generated? I feel like the burden should be on people who are using LLMs.
Yeah it feels like making the tag be “100% Human” or something similar would just make AI normal. As if you can use as much AI as you want and it’s the normal way of coding, which I really hope it doesn’t become.
In the most recent Godot Tomorrow, Emi said that the team is working on AI “rules”.
Why does using AI require a disclaimer but straight jacking someone else’s code doesn’t? Before AI, I used code from Reddit posts, GitHub repositories, and Google when I wasn’t sure how to do something specific. I went thought a process I call “making it mine” where I go through the code line by line, changing variable names, capitalization, curly brace alignment (in other languages) etc. As I do this, I’m reading the code over and over. By the time I’m done, I know what every line is doing. It is no longer code I borrowed from someone else. It is my code, with my signature and my full understanding of what is going on. And it is incorporated into my larger code base in a way that makes sense to my larger code base, not the code it was originally taken from.
I do the same thing with Claude. It suggests code, which is extremely helpful but I don’t just take it at face value. I read it, change it, test it, poke it, prod it, and bend it until it is mine. Claude takes care of my Github commits so it is transparent I am using AI. But I am not producing “AI slop.” I am taking my time and making sure it is right. Should I automatically be disregarded by the community because I use AI as the tool it was intended to be? Nobody had any problem with me using Reddit and Google for the same purpose I am using Claude.
Sorry, I should have been more specific. I was talking about code snippets from tutorials and people explaining on Reddit how to do something specific. I don’t see this as stealing. I am just using the same answer someone else got that had a similar question to mine.
But my original point remains. There is no requirement for “I used Reddit posts to make this addon.” Using AI, in a lot of cases, is basically the same thing. At least from my perspective and the way I use it.
Have you ever actually used AI to code? It’s not what you think. I know this because I use it all the time. Claude is more like a collaborator. To me, I’m the senior software engineer and Claude is my junior software engineer. But Claude works for $20.00 a month which fits my independent developer with no hope of income budget much better than an actual junior software engineer. Claude is great for busy work and prototyping and documentation. When I first started, I made the mistake of thinking Claude could do what I do. It ruined my project to the point that I had to revert to a previous commit. Since then, I have been much more vigilant. It takes someone who knows how to code to use Claude effectively. It is a tool, just like the Godot engine, Blender, Gimp, etc. It has a purpose but that purpose is not what most people think. For me, it is often just a sounding board. I bounce ideas off Claude and Claude either says, “No, that’s stupid,” or “You have a point and here is some more information on this topic you might find interesting.” Or something to that nature.
Not that I am aware of. You would need Gemini, Meshy, or some other AI designed for this specific purpose. I only use Claude to help with code. I plan on hiring human artists for art and music because I can never get AI to do what I want. I use up all of my tokens and don’t have anything usable at the end of it.
I forgot to mention troubleshooting. Most of the time, I love pouring over code and tracking down bugs. That’s the main attraction. But sometimes, I just want to move forward. Maybe I’ve already spent 2 hours banging my head against the wall, trying to figure out a simple problem. Or maybe it’s one of those times when the code is perfect. I’ve checked everything a dozen time, printed out debug information, checked all my variables’ values and logic and everything should just freaking work but it doesn’t. In 30 seconds, Claude tells me exactly where I went wrong. I get back to work in 30 seconds instead of hours or days.
I guess the point I am making is an AI flag is easily misconstrued. It lumps people like me, who are using AI responsibly and producing solid code that has been tested to the nth degree, with people who are typing 3 sentences and distributing whatever AI poops out. I work on my code about 20-30 hours per week. It is months from being usable by anyone else. But once it is complete, I think it will have great value to the community. How many people will disregard Standstill Digital Media addons because they have the same AI tag that AI garbage has?
Just as you have the right to use AI to work on things, other people should have the freedom to decide if they wish to use addons or resources that were made using AI. Not giving people this option is harmful in my opinion.
Besides, there’s a massive difference between looking up a solution online or from a guide, vs using AI code you don’t know the origins of, which CAN have a different license that violates MIT
I think the issue is of a more religious nature. To some of the most anti-AI users, it doesn’t matter if you have used the tools a sound way, just as it wouldn’t matter to many muslims that the “dirty pork” you have used in your cooking is highest quality iberian bellota pork. Some other users in other forums would want you to mark it “coded on a device powered with green electricity”, “coded on a device not made in China”, “coded without woke SJWS involved”, “coded by a non binary” etc etc. You have to decide yourself if you want to play along or not. If someone is really that anti AI they have the option to not use any tools not clearly marked “made without AI” just as people allergic to nuts would not eat some freely available cake if they did not know for sure that it did not contain nuts.
Edit:
Just want to add that I didn’t mean that you should bake a nut cake and hand pieces out for free without informing people it contains nuts! Not a good idea!