Anyone embrace AI yet?

[UniTY]

Old warriors,

I've been using Anthropic's Claude to do so much as of late. It might as well be a cluster of experts working diligently on whatever needs I give it. I would bet it could completely develop just about anything if you spend enough time with it.

Has anyone given it a go against the game resources?

I think a limiting factor is the current method to package game resources/imagery. It's been a long time since I've touched the game. Is there a method to export everything to a standard file type (gif/png)? Would take quite a few tokens, but I would bet you could create a web-based version of MIR 2 (or 3) pretty easily if it's given access to image files it can actually display.

 

[Alecs]

If you read some of the releases around here you'll understand why more datacenters are being built.

Web mir has been tried before and it's possible but it would be a pain in the ass considering the client size, even with wasm.

 

[Jev]

I’ve been using Claude myself and it’s very good but I would advise you check over the code it provides for things like security issues and memory leaks.

 

[NightScare]

I have the £200 a month Codex package, and i smash my weekly allowance in like 3-4 days.

Balls deep in this stuff at work, and at home.

It's quite frankly amazing, ha.


Codex is much more affordable and better than claude though, we use claude at work and opus 4.8 is not great compared to gpt 5.5 on xhigh.


In terms of mir, it can create anything backend you want, but really struggles with anything UI/Image based.

 

[forspinki]

I've been using it for over a year now, i have the £200 cursor plan and £100 codex plan atm and i still go through them super quick. Around 1 year ago the AI still used to make a lot of mistakes and I had to spend a good while debugging but now it adds new features usually the first time without issue. The only thing it still sucks on is consistent image generation with specific specs.

I use a mix of Opus 4.7, GPT 5.5, and Composer 2.5 atm.

 

[UniTY]

I have the £200 a month Codex package, and i smash my weekly allowance in like 3-4 days.

Balls deep in this stuff at work, and at home.

It's quite frankly amazing, ha.


Codex is much more affordable and better than claude though, we use claude at work and opus 4.8 is not great compared to gpt 5.5 on xhigh.


In terms of mir, it can create anything backend you want, but really struggles with anything UI/Image based.

Interesting! Shows how quickly these companies are leap frogging each other. I recently switched to the $100/mo Anthropic plan because I was wearing Opus out in hours, and it seemed at the time most capable. With Fable just released, we'll see it all change again (if you can get more than 3 prompts in without hitting a hard stop due to safety.) That said, I've used Opus to create so much over the last six months. For the sake of it, I setup Claude Code on a remote session on a Linux VM and bought my kid a domain. He's created his entire website, all the games and capabilities, user management everything all by prompts to Opus. Granted, it's on an obscure VM for a reason.. I have no idea how secure his site is lol.

Anyway, I feel like AI could absolutely recreate Mir with little effort. Let it look over Crystal to understand, feed it the game files to extract and convert the imagery over to a format and delivery method conducive for the web.. and probably could have something running in a day. Just wasn't sure if anyone has really gave it a go yet.

Also thought how you could probably feed the imagery through AI to find a way to extrapolate the current files into 3d models. If nothing more, wouldn't have to be 6 FPS running animations..

Just random thoughts!

Post automatically merged: Yesterday at 7:23 PM

Also would be interested in knowing what companies are actually expecting out of AI use these days. In my world, we're actually not allowed to use it.

Post automatically merged: Yesterday at 7:25 PM

If you read some of the releases around here you'll understand why more datacenters are being built.

Web mir has been tried before and it's possible but it would be a pain in the ass considering the client size, even with wasm.

I'd suggest streaming the imagery in from a web source using modern image formats, not some lossless packages

 

[Stavros]

can tell who's using AI especially for bug fixes, as when they release bug fixes back to back, the first lot reappear when the second lot get released

that being said, nothing wrong with using AI, its a very useful tool but in my opinion still needs someone with the knowledge working on the files

 

[NightScare]

Interesting! Shows how quickly these companies are leap frogging each other. I recently switched to the $100/mo Anthropic plan because I was wearing Opus out in hours, and it seemed at the time most capable. With Fable just released, we'll see it all change again (if you can get more than 3 prompts in without hitting a hard stop due to safety.) That said, I've used Opus to create so much over the last six months. For the sake of it, I setup Claude Code on a remote session on a Linux VM and bought my kid a domain. He's created his entire website, all the games and capabilities, user management everything all by prompts to Opus. Granted, it's on an obscure VM for a reason.. I have no idea how secure his site is lol.

Anyway, I feel like AI could absolutely recreate Mir with little effort. Let it look over Crystal to understand, feed it the game files to extract and convert the imagery over to a format and delivery method conducive for the web.. and probably could have something running in a day. Just wasn't sure if anyone has really gave it a go yet.

Also thought how you could probably feed the imagery through AI to find a way to extrapolate the current files into 3d models. If nothing more, wouldn't have to be 6 FPS running animations..

Just random thoughts!

Post automatically merged: Yesterday at 7:23 PM

Also would be interested in knowing what companies are actually expecting out of AI use these days. In my world, we're actually not allowed to use it.

Post automatically merged: Yesterday at 7:25 PM

I'd suggest streaming the imagery in from a web source using modern image formats, not some lossless packages

Ive tried the creating 3D models around a year and half ago, but not tried recently.

It didnt work very well back then.

I work for a gambling tech company, so we are integrating agentic agent everywhere, no one actually hand writes code anymore. Someone actually got fired as they refused to adopt, so its actually crazy the difference between your industry and mine. (Your company will be outpaced by competitors in a year or two if they dont change their tune however).

AI is coming for all our jobs and all industries and you can not put your head under a rock and pretend its not happening.

The frontier models write better code than most developers i've worked with.


The people who have problems or think "its crap" are just feeding ai a prompt and expecting magic, it needs orchestrating and governing properly like anything to be useful.

 

[Alecs]

I work for a gambling tech company

 

[Stavros]

AI is coming for all our jobs and all industries and you can not put your head under a rock and pretend its not happening.

I'd like to see AI dig 2m foundations, lay bricks and pitch a roof

 

[NightScare]

Post automatically merged: Today at 3:07 PM

I'd like to see AI dig 2m foundations, lay bricks and pitch a roof

And this was before ai. Only problem atm is cost and economic scale along with regulation. Give it 10-15 years though.....

 

[Alecs]

View attachment 40036

That's not the flex you think it is.

Anyway.

 

[Stavros]

And this was before ai. Only problem atm is cost and economic scale along with regulation. Give it 10-15 years though.....

View attachment 40037

Thats a prefab, and the picture is heavily rendered. any old machine can build a prefab

Again, would love to see AI try a days work of a tradesman, would fry its circuits

 

[NightScare]

Thats a prefab, and the picture is heavily rendered. any old machine can build a prefab

Again, would love to see AI try a days work of a tradesman, would fry its circuits

Eventually we all be living in prefabs.

Be depressing, but capitalism will demand it, custom homes will be for the rich only, Like petrol cars.

Maybe not in our lifetime, but your children most likely.

You guys are deffo the safest for now, thats for sure.

Anything physical/labour intensive with strict regulation is pretty ai proof for the next decade or so.


Regulations one of the main issues, we have self driving cars but can't get the approval/mass adoption. The UK still has actual tube drivers fgs, and they always striking. (most other countries are fully autonomous)

 

[Stavros]

the world will end before we see AI able to crawl through some of the spaces we have to, cut pipes and tee off of them

the UK will never beable to roll out prefab houses, unless somehow the cities get flattened and rebuilt. Even then, building regulations and control will have a field day halting all those plans

 

[knives]

I have the £200 a month Codex package, and i smash my weekly allowance in like 3-4 days.

Balls deep in this stuff at work, and at home.

It's quite frankly amazing, ha.


Codex is much more affordable and better than claude though, we use claude at work and opus 4.8 is not great compared to gpt 5.5 on xhigh.


In terms of mir, it can create anything backend you want, but really struggles with anything UI/Image based.

you need to look at implementing guard rails (AGENTS.md / AGENTS.overide.md) to reduce token usage.

simple things such as getting it to have its own inhouse council to look at things with an alternative perspective will see a improved output and token reduction aswell

 

[Jev]

you need to look at implementing guard rails (AGENTS.md / AGENTS.overide.md) to reduce token usage.

simple things such as getting it to have its own inhouse council to look at things with an alternative perspective will see a improved output and token reduction aswell

This is my issue, I need to create a init.md to give some params to lower my token use.

Obvisuouly i'm opening claude directly into the source code's folder but sometimes i need it to go elsewhere (i normaly give it the full path for the file) but i imagine init.md may help it not go awol and search more places it needs to.