How far are we from AI recreating EQ99?

[Dulu]

https://www.youtube.com/watch?v=MWl2qQRyutc

Watched this video, where the guy basically recreates Doom in an hour.

Keep in mind, Doom is a single-player game, and came out in 1993.

But... the video is 3 months old and was probably recorded some time before that.

So the question is, how far are we from making a game like P99 in AI?

[DeathsSilkyMist]

Originally Posted by Dulu [You must be logged in to view images. Log in or Register.]

https://www.youtube.com/watch?v=MWl2qQRyutc

Watched this video, where the guy basically recreates Doom in an hour.

Keep in mind, Doom is a single-player game, and came out in 1993.

But... the video is 3 months old and was probably recorded some time before that.

So the question is, how far are we from making a game like P99 in AI?

Even an older MMO like Everquest is significantly more complicated to produce when compared to Doom. You can already download an open source Unreal 5 multiplayer shooter template, so there's no real need to remake the wheel in a probably worse way with AI.

Complex games like Everquest (and beyond) are going to be one of the last frontiers when it comes to AI generated content, especially multiplayer games. Vibe coding networking code is usually the easiest way to open up security flaws and cheating. You still need experienced programmers at the wheel to check AI's work at all times.

[BradZax]

I saw a youtube video about putting people on a moving boat in an MMO style game and it looked like he was building a nuclear bomb so I think yeah we're a ways away still.

Originally Posted by DeathsSilkyMist [You must be logged in to view images. Log in or Register.]

You still need experienced programmers at the wheel to check AI's work at all times.

But how long until AI turns each experienced programmer into the equivalent of 50 experienced programmers [You must be logged in to view images. Log in or Register.]

[DeathsSilkyMist]

Originally Posted by BradZax [You must be logged in to view images. Log in or Register.]

But how long until AI turns each experienced programmer into the equivalent of 50 experienced programmers [You must be logged in to view images. Log in or Register.]

This is true. AI can act as a force multiplier for experienced programmers.

The main concern I have with that thinking is this will probably lead to hiring less junior programmers.

The question is: Is that a good thing? The only way to create experienced programmers is to train junior programmers. Until AI can do everything for us, we don't want to stop training people.

Ideally you keep hiring junior programmers, and you use AI to increase productivity.

[BradZax]

Originally Posted by DeathsSilkyMist [You must be logged in to view images. Log in or Register.]

The main concern I have with that thinking is this will probably lead to hiring less junior programmers.

Maybe, or glass is half full take:

Jr programmers (who have the most up to date knowledge of cutting edge tools because they are young and adaptive) will be able to compete with companies that wont hire them, and decimate them in the market.

Ye old "disruption" technique [You must be logged in to view images. Log in or Register.]

[Swish]

Originally Posted by DeathsSilkyMist [You must be logged in to view images. Log in or Register.]

Until AI can do everything for us, we don't want to stop training people.

You don't want this.

[Dgc2002]

Originally Posted by Dulu [You must be logged in to view images. Log in or Register.]

Watched this video, where the guy basically recreates Doom in an hour.

That's a CRAZY representation of the video.

The 'better' version he made using grok had:
The most basic of WASD movement
Shooting is just point and click
Enemy behavior is

• move towards player position at a constant speed if they're not close enough
• face player
• shoot every n seconds

The shooting every n seconds portion wasn't even finished within the hour time limit but then worked randomly after adding the imp.

I wouldn't be surprised if you can just find a free boilerplate project that covers all of that and more as a starting point.

[loramin]

OP the short answer is that we're a longer ways off than you think.

I code with AI daily, and it's incredibly powerful. I can say "go do this task", and it does it. Most of the time it even gets it right.

However there are two major issues. First, the context window. The AI can only "remember" so much at a time, out of everything it reads, everything you tell it etc. The context limits are inherent to how AI works: until there's some new innovation occurs, this isn't going to get much better.

The second problem ... last night I asked it to fix a test, and it tried to change the test to check the wrong thing (so it passed, but the issue wasn't fixed [You must be logged in to view images. Log in or Register.]). AIs can't (really) understand what's "right", as they just know what their linguistic model predicts.

It's like that old joke about the repair man and the hammer. AI is a powerful tool, but it doesn't fix things, the human intellect still does.

A repairman was hired to repair a large machine in a factory. He showed up, examined the machine, then tapped it once with a hammer. It started up. The factory owner was pleased, but not when he got a bill from the repairman for $100. He thought that was outrageous, and he asked for an itemized bill. So the repairman handed him a bill which said:

Tapping machine with hammer: $1

Knowing where to tap: $99

(Re)creating real/serious games still requires humans because of the above ... just not as many as it used to.

[Kich867]

Having started using it more at work, also programming, it's not bad at certain things but anything that requires any sort of context its often dumb as hell at. Or, to reinforce loramin's point, like I had it help me spin up some AWS infrastructure, it organized it and wrote it well but it included options that don't exist on those resources / tried to use conflicting options / just had general bugs in it. So it requires extensive review.

It helped me because it wrote up like 900 lines of cloudformation template, but the fiddling afterward to correct it was also time consuming. I'm skeptical on whether in the end it actually did save me any time from the amount I have to baby it.

I think it was MIT that recently did a study on this that: despite the project managers claiming AI is improving speed, and despite the devs themselves believing that AI is improving their workflow, when you actually observe and time how long tasks take to be done, tasks with AI are being completed about 20% slower.

(IIRC they did a study with various open source developers, found similar tasks, had them do tasks with AI and tasks without AI, and they were consistently faster without it)

[BradZax]

Originally Posted by loramin [You must be logged in to view images. Log in or Register.]

First, the context window. The AI can only "remember" so much at a time, out of everything it reads, everything you tell it etc. The context limits are inherent to how AI works: until there's some new innovation occurs, this isn't going to get much better.

I think we're just waiting on more power. For example. Google image creation with Gemini using context costs about 3-5$ in wattage to generate a 6-10 second video. Vs their diffusion model for video generation is more like 5 cents or less.

That doesn't mean anything except it just gives you an idea about how much wattage context adds to the equation!

I am hoping corporations using their own nuclear reactors will be give us a big jump in usage.

https://www.latitudemedia.com/news/o...-nuclear-push/