Not seeing AI as a tool?

General
1

I’ve waded through this community for months and found a lot of people that are helpful and knowledgeable that are posting and responding in detail all the time that I learned from as well.

But there is this certain feeling around AI forum posts that act like AI is this unknowable mystery. It is either:

  • Really Good at everything and should be 100% be trusted and used
  • Really Bad at everything and should never be trusted and used

When looking at AI for programming there is a few things I keep in mind when using it:

  • I always assume it is wrong. This helps keep me looking at the code and not assuming everything will be fine.

  • I only use it for ideas in the code base. To be specific I mean on where to go next in the code base I built myself to reach a certain goal also made by myself, not generated by Mr. AI and I mostly ignore what suggestions it suggest and go with my planner instead.

  • I then look at the systems in the code the AI touches with the code it generated to get a better understanding of the thing I created and quickly become familiar with it.

  • I then refine my code base to be easier to work in seeing the flaws in my initial structure design.

At no point I am copying and pasting giant code block and not understanding what is happening. I only use it for refining what I have created.

This is what AI is good at. Trust me when I say AI is not good at generating your code for you. It is good at understanding how the Lego fits together and showing you how it is put together.

Why the AI could feel like a genius because it built something you didn’t understand sometimes, I know there have been moments where what it generated doesn’t work at all that people choose to ignore. (If you haven’t felt this you haven’t made anything big enough with AI to crack).

AI is good and bad at things you have to notice and be aware of where it is, there is a middle ground where you can be a better programmer, just don’t praise it like it is a god; ALSO don’t dismiss it like it’s useless. It is a tool.

(p.s.

I wrote this by hand and then asked Gemini this question: “what are keypoints that I would need to refine to make it more concise”

It praised me at first without giving actual feedback a really bad problem

I asked: “So nothing really needs refinement your just glazing me“

Then it gave me feedback I could work on (remember assume its always wrong)

but the feedback it gave dismisses my logic that I have noticed AI in general get wrong and I noticed the model is trying to not let itself look bad…. wtf google :angry:

I then used a uncensored local model and using LMstudio asked “what are keypoints that I would need to refine to make it more concise“

It worked so much better at telling me writing tips and teaching me actual writing concepts. *Even though I reset context and model every 3 or 4 chat and responses because it becomes too praise heavy otherwise.

Still written by hand.

)

Rough Draft

I’ve waded through this forum for many months and found a lot of people that are helpful and knowledgeable at responding that I learned a lot of things for myself that make me better at programing.

I’ve also noticed that posts about AI get a lot more traffic than other posts because it is the current hot topic for computing and has gone mainstream.

But there is this certain feeling around AI forum posts that act like AI is this mystical thing that we don’t control that it is out of our reach and we can’t understand what it does, and it only exists to be bad or only exists to be good. (2 extremes)

When looking at AI for programming there is a few things I always keep in mind when using it:

  • I always assume it is wrong. This helps keep me looking at the code and not assuming everything will be fine.
  • I only use it for ideas, to be specific I mean on where to go next in the code base I built myself to reach a certain goal also made by myself, not generated by Mr. AI
  • I then look at the systems in the code the AI touches with the code it generated to get a better understanding of the beast I created and to be more quickly familiar with the code base I am in.
  • I then refine my code base to be easier to work in seeing the flaws in my structure design
  • At no point I am copying and pasting giant code block and not understanding what is happening. I only use it for refining what I have.

This is what AI is good at. Trust me when I say AI is not good at generating or refining your code for you.

It is good at understanding how the Lego fits together and showing you how it is put together.

Why the AI could feel like a genius because it built something you didn’t understand sometimes, I know there have been moments where what it generated doesn’t work at all that people choose to ignore. (If you haven’t felt this you haven’t made anything big enough with AI to crack).

AI is good at time, it is bad at times, there is a middle ground where you can be a better programmer, just don’t dismiss this tool like it is the plague and don’t praise it like it is a god. It is a tool.

(p.s.

I wrote this by hand and then asked Gemini this question: “what are keypoints that I would need to refine to make it more concise”

It praised me at first without giving actual feedback a really bad problem

I asked: “So nothing really needs refinement your just glazing me“

Then it gave me feedback I could work on (remember assume its always wrong)

but the feedback it gave dismisses my logic that I have noticed AI in general get wrong and I noticed the model is trying to not let itself look bad…. wtf google :angry:

I then used a uncensored local model and using LMstudio asked “what are keypoints that I would need to refine to make it more concise“

It worked so much better at telling me writing tips and teaching me actual writing concepts.

Still written by hand.

)

2 Likes
2

The environmental and societal impacts of how these LLMs are distributed and bolstered by venture capital rug pullers should be enough of an argument to dissuade any usage by anyone.

As a tools programmer and teacher LLMs are the worst tool in a historically awe inspiring and amazing community of programmers, we have a true Library of Alexandria. Where our culture is open-source making anything for others to improve freely, meaning in both libre and financially. OpenAI and friends have wedged themselves in the middle disrupting our free licenses, destroying search engines for exploring resources, all hoping to shackle your work and brain to their per-token subscription.

“AI” is a tool to bastardize programming, it un-democratizes what we’ve built. It’s ruining the world around us, and gouging prices for anyone poorly fated to live on the same planet, worse yet town or city.

I am very glad you’ve liked this forum, I am very proud to be a part of it. I hope others can find it a useful resource too, I can hold this hope because it is open and free.

10 Likes
3

Sorry for this not being on topic but holy shit I swear to god I’m looking forward to the day ai topics die out on this forum. It’s too much. It’s just too much nowadays. Ai this AI that. Every topic is AI. I go onto general as a category and all of its AI. There’s so much AI. Not to take away from good AI questions. Not all questions about AI are bad. But every single AI post spirals and holy god it’s too god damn much.

5 Likes
4

Does that warrant having all conversation touching to AI on this forum to be turned a place to voice opinions on AI when all they had were a question on how to do things or a opinion on something someone made?

What can we do to about how LLMs are expanding and integrating into our society other than a country or local governments stepping in and setting boundaries like blocking all online AI related tools to schools district wide or holding accountable the companies who are training the data on things they don’t hold the copyright or license to?

That is a huge problem, yes, I agree but when it seeps into the forum it becomes harder to read and stops discussion about the original topics.

The point I am trying to make with this post is to help people realize and not start that kind of conversation in this space, I love learning from people but when the conversation turns into a what AI is, in a place about Godot and making games made me, want to vent a little which is why I made this post.

5

I feel all the hate towards AI is justified. Especially on this forum.

Many people here have likely been affected by AI in some way. Maybe it was their job, maybe something smaller like RAM prices but it’s a real life issue.

That being said as you see at the top the AI stuff on the forum is just too much now. I always like putting down stuff like blured. But at the same time so often it’s just a pain.

I personally am against the use of AI in programming for so many reasons. Don’t get me wrong it has uses. I won’t deny it can be useful.

I just think using it goes against so much.

I would say of all areas in programming game Dev is the closest to being a form of art. So to me using AI for your game is like using AI for your art. Id say that’s the best analogy for how I feel about AI in game Dev.

7 Likes
6

I would say so. Superoptimally if there is a conversation about LLMs nobody would reply and it would fade. I usually mute these topics.

If someone is looking for help then I am happy to help them, I think it’s fine if others do not wish to help from a LLM foundation code base, but I try. It can display one cares more about having a game made than the academic pursuit, the latter is far more rewarding for volunteers and easier to help.

I usually see AI-centric posts and death spirals happening in General where the main topic is about AI or an AI-generated thing; I absolutely see that as a valid through point to voice any concerns about AI. Creators should know where potential customers, clients, and community stand on their product.

If these spirals happen in Help then I am sorry, I would agree it’s best to try to help, point out flaws, no tirades and dog-piles. I do my best to ignore AI tells and at worst point out how the AI fails in the original post.

4 Likes
7

I agree with all your point except about using LLMS for coding.

I feel people can use coding LLM’s that is in a way harmful to themselves and as a whole by not making decisions themselves and starting to rely on the LLM too much.

They use it and say they can write 10,000 lines a day which is bull, and they say there is no more need for human programmers, again, bull.

I feel that when majority of coders will look and see what is happening in the world, not the internet, and see that AI will not replace them, corporation will, and try it as a tool.

It does help with coding a lot more than you think.

I wish AI diffusion art was never invented, it takes too much away from actual art.

I wish someone would control the corporation and help all these prices settles instead of letting them run wild building and taking electricity where ever.

1 Like
8

Yeah I’m not saying you shouldn’t use AI to code at all. I personally hate the thought of it but feel free to do it on your own project even if I can’t recommend it.

But to me when someone tries to implement AI in certain ways it hurts the artistic part of game Dev. For example an AI engine where everything is optimised so you can write a ton of prompts. That is the point where I draw a line and it’s something I don’t want to exist in this community.

It’s really mainly about that artistic part. Cause this isn’t just coding, it’s game Dev and to me there is a big difference.

I hope that makes sense, I’m writing this sleep deprived as hell lol.

5 Likes
9

I think tbh that thinking this fails to understand the effect it has on the skills of the user, the atrophy, but also I honestly feel that people who think it helps that much don’t know enough to judge it

4 Likes
10

To be honest all the LLM code I’ve seen has been derivative at best and crap at worst, I’ve seen the repos with all LLM code and automated PRs and it’s slop from floor to ceiling

5 Likes
11

That true. If you don’t understand what a system(code) architecture is you would wouldn’t even think of the LLM being wrong you just ask the LLM to try again and it would do bad system architecture decisions, patch it, then move on.

You would need a baseline knowledge before integrating it within your workflow.

Which some people take for granted and just let the LLM do it. Guess that why there are so many bad apps coming out with bad security on head line news.

12

Exactly, from what I’ve seen people who actually know how to write the relevant code find it slows them down and requires far too much work to fix it, the people who it “helps” the most are those who don’t know the code is bad

2 Likes
13

The only thing I use AI for is copy-pasting my code in my project, and asking if it can tell me what it does. If it does so correctly, that means the code is probably easy to read and clean. If it is incorrect, I need to polish the code some more.

14

I would ask to try and make it yourself first and see if your can get it to work by yourself then ask the AI for bugs.

I feel this would help you understand code more and rely on the LLM less and make you a better programmer in the process.

When it tells you a bug it makes you better at understanding what part of your own logic was wrong and not relying on the LLM to create the logic for you (coding is 95% a logic puzzle).

But if you change the logic bug from the LLM and the game responds in a weird way than what you were expecting than you can know that the LLM was wrong but understand now have a understanding what to look around for.

15

Given that LLMs fail to do this when writing its own code I wouldn’t trust it to proofread yours, and knowing when it responds in a weird way is difficult, some bugs are very unintuitive

1 Like
16

The biggest missing thing in your post IMO, is that you clearly have some programming experience, and do not disclose how much. All the things you suggest to do when using LLMs require programming knowledge. The average person coming here for help who uses it as a tool does not have that knowledge and therefore cannot follow your instructions. Perhaps I’m not understanding the target audience of your post however.

You also use the terms LLM and AI interchangeably. Which is fine if you’re selling them. I think it behooves us as game developers to use the term LLM when referring to LLMs, as AI has a meaning in game development already.

This just seems like you’re policing other people’s feelings. “You can feel that way, just please stop posting about it.” No one is asking you stop posting about defending LLMs. How is it then appropriate to ask people to stop chiming in with their own thoughts?

That’s an interesting take. And I’ve seen you try to help people I’ve not engaged with or abandoned. I just find it frustrating when people want explanations of things that are beyond their grasp to understand because they don’t know how they got there to begin with. Props to you for doing that.

I agree that we should be kind, but I also think it’s appropriate to warn people that this LLM-generated problem they’re encountering in their player movement code is just the tip of the iceberg.

OMFG No. I spent the last week debugging AI slop that one of my employers made. In fact, I had to have a conversation with him about the speed in which he did about 20 PRs with Claude while I was still trying to debug the first one. It’s an F-ing time sink if you actually know what you’re doing. A horrible, nightmare-inducing suck of time.

Today I sent him the code I fixed, and he ran it through the LLM. It praised my programming skills, and then said that the PR shouldn’t be accepted until I fixed three potential bugs that it had created!

Some of my own thoughts on the topic.

I’ve been using LLMs a long time. Longer than ChatGPT has been around. They’re amazing for some things, and they are a really neat piece of technology. They have been advertised as the solution to many problems that they cannot solve.

We also haven’t seen the end of their impacts on society. Factory workers in India are wearing cameras to train the robots replacing them. And even more worrisome is some of those workers are college graduates who cannot get a job with their degrees so they’re doing this to make ends meet. Meta is going to start tracking employee’s keystrokes, clicks, mouse movements and take screenshots of their computers to train LLMs. Even though they never agreed to train their LLM replacements when they were hired.

I talked a few months ago about how I was using Claude AI for simple tasks and it did pretty good. After the mass exodus of people from ChatGPT to Claude, Anthropic has thrown up a much stricter paywall. I used to be able to ask questions for an hour or so and then have to wait 2 hours for more bandwidth. Now they’ve reduced the models I have access to, and I get one question every 5 hours. This means that it’s unusable to me unless I want to pay for it.

I think we will see more of this and LLMs will stop becoming so popular because the paywalls will go up, and they will continually get more expensive. A company that is pissing away millions or billions cannot survive forever. So this may be an enjoy it while you can scenario. In a few years you may not be able to afford using an LLM to make your game.

Whether you intend to or not, you are supporting these changes.

6 Likes
17

In my experience I feel AI works better when working on code given to it, not so better when generating it.

18

See above. Not true in my professional experience.

4 Likes
19

Please all remember the following: programming takes logic and reasoning, two things an LLM cannot do

They are language models, they don’t think, and they “trick” you into thinking they can because they can talk

4 Likes
20

I had no idea models were being trained in that way holy shit, thats horrible

I use local models 9B parameters 99% of the time and I only use it on code I wrote myself like in the method describe originally.

From this post I was hoping people would take it as OH I can use AI at a small area within a code base and figure out how that works faster when already understanding the language and code base hope someone could learn from it. Maybe I should of worded my post more positively and not use AI in the title and not come in swinging (I didn’t mean it to seem that way) or maybe I should of put it as advice for people who know how coding works in general and use AI specificly. Which is why I had a angry without realizing tone my bad.

I didn’t want the conversation to go this way, sorry I must of worded it too strongly @gertkeno

For some reason I always used them interchangeably somehow I didn’t realize that, LLM is the coding one and AI is the general

Sorry I really like privacy. Sorry if it comes off as negative.