Using AI assistance. How much is too much?

General
1

With there being discussion on the use of AI to develop Amstrad games where do you draw the line? (draw the line as in the sentence ‘draw a line in the sand’ to say how much you will accept but no more)

Using an online AI assistant to help find documentation (e.g. the online AI assistant in google search)?

Using AI to discuss ways to do it but then coding it yourself? (e.g. discuss with ChatGPT how something could be done)

Using AI to develop a small or large tool which is then used by you to develop a game where AI has no further involvement (tool could be to make a dsk, cdt, cpr, a graphics editor of some form, a music/sfx editor, a ‘make’ toolchain, some kind of tool to help accelerate development)?

Using AI to generate ‘boilerplate’ framework code which goes into the game (e.g setup hardware, setup a dummy interrupt, an empty main loop) which then helps you to focus on coding the actual game?

Using AI to develop a prototype to test something, AI has coded quite a lot, the idea works out, now you are happy, understand it and you re-write it yourself?

Using AI to debug code only you wrote but you are struggling to find the problem after hours of looking.

Using AI to perform some conversion, e.g. graphics, code flow (maybe from another computer and maybe another cpu), but you then you are providing the remaining code, debugging and fixing code the AI generated data and tweaking it and polishing it etc. AI has some involvement but ultimately you have most involvement?

You can code, maybe partly or well, but you use AI to do various bits, you then complete it and ask others to help with graphics and sound to make a great game.

Using AI a lot but because you can’t code that well (or at all) but really want to see your game idea on the computer you love and you instruct it on what to make and you instruct it to polish it and make it great to play? (vibe coding essentially). AI then does majority of the work including generating some graphics and some sound and the gameplay. (Your idea may be unique, or it may be similar to others out there - does that make a difference?)

You can’t code, use AI to make a game idea you had, there are bugs and you’ve asked the AI to fix them but they are still there and you ship the game with those bugs.

Then how important is it that you are told AI assistance has been used and how much has been used and finally does it matter if it’s free or paid for.

It is interesting to read people’s thoughts.

2 Likes
2

I’ve thought long and hard about this over the last year or so, because it’s clearly been on the rise and though my initial thoughts were “it’s all a fad” - I can certainly see that as the tooling improves, so can the results (potentially).

And so I’ve come to the conclusion that, for me at least, it doesn’t matter at all. Yes, there is inevitably going to be a period of badly rushed out slop from people suddenly enabled in game development in a way they’ve never been before. We’ve seen the same thing when game maker tools get released, or things like BASIC compilers that simplify production in any way.

I chose instead to focus on the end result. If someone makes a game with AI by just minimal prompting, the chances are the end result will be exceptionally low quality. And I will either ignore it or call it out as low quality, depending on what seems appropriate at the time just as if they’d coded every line by hand (I’m far more likely to be lenient to someone obviously learning and not trying to make money/get YouTube hits/spam the forum etc)

Over time I expect to see a lot more higher quality efforts though and that’s something I find deeply interesting. Getting AI to do what you want, and only what you want, is a skill akin to managing a team of junior developers. And the potential now exists for any one of us to essentially direct our own teams with far more capacity than even the big guns like Ocean would’ve put on CPC development back in the day. What might the results look like? I suspect we will start to find out over time. They are interesting times indeed.

7 Likes
3

Once your at the point of not understanding what its doing, then, thats it, because your at the whim of it.

My experience of using AI was with DirectX12, using Co-Pilot (shock horror), but it was great because when converting a DirectX9 library I had, it taught me how to take that, and modernise the engine into DirectX12, it explained DX12 devices plus swap chains, command queues, command allocators, command lists, descriptor heaps and tons more stuff. It really taught me a lot, and thats what I based my runtime around, seriously, DirectX12 is deep. And if I didnt understand something, I’d ask it to lay it out more. It worked a treat.

I just used the free account of Co-Pilot for all the advice, and just recently, Microsoft have moved the goal posts, usage on the free account has been throttled, seriously, it’s went from being a really handy tool (For me), to be about as handy to have when I want to know when I need to prune my apple trees.

But, my usage for it is done really, so I’m not disappointed really, because I used it to learn, not write stuff for me, and thats the thing… If people use it to code, actually to write code for them, they are going to be trapped into a paywall, it’s getting there, and these companies will only charge more and more… They own the data you have technically.

So far me, use it to learn, get it to explain everything, not just bang out code and all you do is build and run… It’ll probably bite you somewhere along the lines.

Dabzy

4 Likes
4

I don’t use LLMs for my personal projects. My enjoyment of it is the process of making it myself, so it doesn’t make sense for me to get an AI to do what makes me happy.

I hasn’t even crossed my mind.

7 Likes
5

I wonder about the players what their personal line in the sand is.

When it comes to modern games features like DLAA, DLSS have used AI in their development. So even a game written without AI that uses those features has been touched by AI.

When it comes to slop maybe we kind of have that in games and films before AI. What about Blaby Game’s Deathkick?? Or many clones of pacman and city bomber?

I know there is more to it when it comes to AI.

Personally I see the benefit of assisting to do something that I would not find exciting, setting up build environment, setting up boilerplate so I can do the exciting bits. I see both sides, I’ve used AI coding to assist at work. I can see good and bad. I can also see the cost getting more expensive over the next year or two.

2 Likes
6

TL;DR: to me only the quality of the result matters. If you accept using a C compiler, a music or graphics editor on a modern machine, AI is just another (powerful) cog in the modern toolset.

Thanks @Arnoldemu for raising this question. It’s really a matter of current interest.

IMHO this is only another layer on top of a cake that has several already.

Consider different stances:

Let’s start with the opinion that using any technology that was not available (or maybe even “available but too expensive”) during the CPC lifespan would be “akin to cheating”. You’d develop on the CPC itself (a real CPC logically).

Then you can allow more using modern technologies:

  • developing on a CPC emulator, perhaps running at higher speed at some steps – no qualitative change, just speed.
  • creating graphics or music with the comfort of a modern tool
  • using version control e.g. git, or external project management tools
  • using a higher-level language, typical example C that only a modern machine can compile to efficient Z80 code e.g. with SDCC
  • finally use AI to help you do, or even do itself any of the above (game design, idea, graphics, maps, music, code).

Each of items above is a layer. To me AI is just another layer or tool. If any of the previous layers feels like a problem to you, then AI probably will, too. Conversely, if the previous layers felt okay, then AI will probably.

What matters to me: good result that works on a real CPC

As already explained on Portability factor as fraction of all CPCs that could run a prod., quoting:

If it could have been a widely run prod during the era, then it’s totally fine: any tape-based prod or disc-based prod that would run on 464, 664 and 6218, green and color screen is 100% Legit.

I’m fully okay with using any modern tool, including AI, to create a prod.

If the prod is weak, buggy, it’s a bad prod. Whether, at any step, you used AI to make it or not is irrelevant.

That covers things like:

So, “used AI to make it” is not a criterion. Still, “a screenshot that look like it was made with a single ChatGPT prompt, hence AI slop feeling” is a quality judgment (that happens to be correlated with the use of AI).

Plot twist

Now that the “wow” effect of writing a prompt and getting an image (without much actual control) has faded… we realize than creating great prods using AI still takes time and efforts.

3 Likes
7

I’m sorry but I can’t agree with that. Current iterations of LLMs are not “just another tool”.

From stolen data used for training to environmental impact, or the effect that is having in society such as deskilling or heath and safety concerns; you can’t compare it to using a C compiler, a image editor or a modern machine to make CPC games.

May be your TL;DR is losing any nuance in the point you’re trying to make.

6 Likes
8

I was going to post something similar, and even if you put aside the stolen content and environmental impact, AI is completely different to a C compiler or an image editor.

Your C compiler can’t create code from a prompt, you need to know the syntax etc.

And similarly an image editor will only create graphics as good as your talent is, it won’t make images based on what you type that you want.

And for me, that step of having to learn coding (or even using a game creation tool) requires some dedication and is of more value to me than a game that has been made by AI from a series of prompts (even if the human has spent time telling and re-telling the AI what they want).

6 Likes
9

Thanks @reidrac and @slider271 for your insightful replies.

It’s a fact that AI generates images based on what you type, but are they good? The usual signs of efforts (e.g. intricate details) tend to become repellent signs of AI-generated slop. For an example see an image of a burrito (image found in a LA times article with section title “Your AI agent has no sense of taste”).

Definitely. It’s hard or even sometimes not possible to convey a precise meaning in a few words. Refining the wording is more effort for the writer (whoever reads), and more words are more efforts to a human reader.

On the CPC scene

We may expect some quick-made prods but in some sense the basic economic mechanism might survive: good prods still require efforts, and we may have some even better prods than before.

This would need to be refined: full-screen graphics are a possible aspect. Can AI generate good sprites, palettes or some of the many other things? Maybe.

Can AI create great prods without effort? Given the testimonies I had “My ambitious project started well then the AI and I got swamped in bugs and regressions.” I doubt.

One data point

Very interesting data point about the current state of AI thanks to @TITAN AmstradMuseum TITAN : Pendu AI. My opinion: The game itself is working, I see no bug. The colors and all lack taste. in its category of BASIC-based game without too much ambition, “Pendu AI” is on par with 1985 type-ins in BASIC, which tend to not fully use the machine capabilities. For a sample of type-ins see Tilt Numéro HS 01 - 4ème trimestre 1985.

Pendu AI:

pendu AI screenshot
pendu AI screenshot768×536 1.72 KB

A 1985 type-in that uses the Amstrad CPC BASIC capabilities:

1985 type-in 'cata-tir' screenshot
1985 type-in 'cata-tir' screenshot768×536 2.25 KB

Collateral damage

I realize your reply introduces an aspect often discussed in general but not yet in this thread: collateral damage.

I focused on “That new powerful thing may help us write better CPC prods (or waste time trying), is that okay?” which is a different concern from “Should we all ban that AI monstrosity which may destroy our civilisation?”.

10

I’m not proposing a ban. What I’m saying is that your “is just a tool” comparing it with other tools that don’t have the same, as you well put it, collateral damage, is simply wrong.

Which is not terrible, but not everybody has considered all these things, so let’s make this thread informative.

@Arnoldemu I didn’t share my opinion as a player! Currently, it generally puts me off for some of the reasons I shared. Is that true for all cases? Probably not. If someone used AI to assist in some less visible aspects of the game (because the worst offender is generally the art), that could be different.

I personally like that the use of AI is disclosed. Why not? Do people feel ashamed or consider that their output is lesser because of it?

Then let the players decide what is important for them.

4 Likes
11

For gamers I wanted to show there are different levels of AI use and not every AI assisted game is slop. For coders we know the areas where AI could assist if we chose to.

Take:

and

Both are ports both used AI and openly said they have and both are amazing.

I am concerned about the impact of AI from environmental to economic.

I am happy to read points from coders and gamers.

12

I have always been, or at least since as early as I can recall, as much interested in the craft that went in to making a game as much as the game itself, and this is only more relevant in a world virtually drowning in games. So whether as a developer or player, for me the answer is ‘no AI at all’. I don’t care how ‘amazing’ something made with AI is, it is of no interest to me.

5 Likes
13

So having played Super CPC Tennis, I’ve changed my thoughts a little.

I still fundamentally disagree with using AI to vibe code games, especially where the author has little or no idea how to read or debug that code. However, in this case, where the game is a direct code translation from Gameboy to CPC (with some tweaks to use CPC+sprites) I do admit that I’m less against it.

This does feel as though I’m being hypocritical, as I still believe AI generally isn’t a good thing for our environment or for humanity in general, but in cases like this, where AI isn’t using its stolen knowledge to copy existing art, I’m less concerned. Although obviously it is using copyrighted Nintendo code to create the CPC code, but the dev is releasing it for free…

OK, I’m conflicted. I kind of feel where something is created that wouldn’t have been done by humans because, basically it’s too boring/uninteresting (recreating the Z80 code from the Gameboy to the CPC without improvements) that isn’t terrible, and I would hope the dev did actually create the Plus sprites themselves (though I’m guessing probably not :frowning: ). I don’t know, I did enjoy playing the game a lot more than other AI created projects I’ve tried so maybe that’s the difference.

I’m going to have to think about this!

2 Likes
14

If AI is used as a tool to support development, I’m pretty sure that can help a lot and I don’t see reason to avoid them. Or stuff like a 1:1 conversion can definitely benefit from AI assistance.

But I also have the impression that we see a growing number of “sloppy” pure AI games. They can even be based on good ideas and are not necessarily bad but they lack the final polishing that a game developer usually puts into its production - and if you lack the skills itself, you can’t do that after your favourite AI tool has done all of the coding.

My personal perception of the AI games is, that if AI is doing almost everything, except for the idea the result is usually just “meh”.

I would really appreciate if those “pure AI” productions would get their own category on CPC-Power (and similar) or even a sub-forum here. I would prefer to know if I praise real creativity - or prompting skills.

4 Likes
15

That got me thinking, the forum also includes a MCP server, so conceptually it can end up being

“Hey google, come up with an idea for a CPC game. Develop the game. Test it to make sure it runs. Make a dsk with it. Connect to the cpcwiki forum and make a thread about it” :smiley:

3 Likes
16

Then it could just play the game for us as well…

1 Like
17

That reminded me of Norman Spinrad’s book, Journals of the Plague Years. In it, people would use robotic prosthetics to perform sex -I don’t remember the reason. And in a scene of petty revenge, a girl, rather than having sex with a boy, has the devices perform sex in front of them instead.

We’re getting there.

2 Likes
18

I am using AI quite much at work nowadays and can’t imagine going back, tbh. It’s just so powerful, spares me hours of searching, reading (boring or badly written) docs and struggling with the Alt key to get any of “{}|[]” and the likes. :laughing: It’s soooo slow that way. But I don’t just write code, but also e.g. create boards in Grafana, work with CI, k8s, many technologies we integrate. Nobody can know all of this by heart, so much work involved searching or reading docs. I am so glad that this is mostly gone. I get things done. And things explained. And I think my learning speed has improved because I often ask about the hows and whys. There are many things about coding on modern systems in modern languages (mostly C++ and Python in my case) that I do not love.

On the CPC, on the other hand, I can’t imagine using AI to write Z80 code for me. There’s something intensely satisfying about it, squeezing a couple of NOPs or some bytes out of a piece of code. Being in full control of everything. I don’t think AI can do that (yet) and I want to do this myself. That’s what I love about coding on the CPC. I use AI to do research though. It basically replaced search engines and sites like stack overflow or special interest forums.

5 Likes
19

One noteworthy chat I recently had with ChatGPT was about C++ value categories. The official docs and most things you find online are just incredibly opaque and horribly worded. I was never really sure what to take out of it. Then I asked ChatGPT it to explain it using Z80 instructions for working with memory. Everything just clicked and I finally understood.

I still wonder why the official docs are written in Immanuel Kant style, though.

1 Like
20

Anything STEM-related I just cannot believe the time saving not having to find an appropriate reference to clarify a technical point. I really hope younger generations seize this opportunity!

CPC creations though, I’m pretty sure for most people the end product isn’t the motivation so I’m not sure why you’d use it.

1 Like