Making games in Go: 3 months without LLMs vs. 3 days with LLMs

We absolutely did see an explosion of good games since the 2010s. I won’t argue if it has anything to do with the proliferation of accesible game engines, but if you spend any time in the space you’ll know there’s now tons of games each year that have years of polish yet get little attention. Was pretty different back when Indie game: The movie came out.

And the main reason why actually making a game with interesting gameplay is more relevant than discussing what is the best language to do a game on.

My litmus test for generative AI: generate a complete spritesheet for a 2D pixel art action game, e.g. only for the battle tank or main hero movements. No success so far.

As a much lower bar, I'd love it if more of the LLMs used for coding were actually multimodal, accepting images of games as part of the context. In general, LLMs are far better at generating web apps than they are at pure 2D or 3D games that use graphical APIs. There's far less training data, and there's no way to test anywhere close to what you can do with testing on a DOM. In an interactive game, with physics, animations and game logic, the AI just falls on its face because of the complexity.

The front end is still react. But I'd be curious to know if LLM's are less prone to errors generating code for strongly typed languages over others like Javascript.

The part of this that always confuses me is like nobody's aware gamejams exist, this has been a thing long before the LLM craze and people have been producing decent games on very limited timespans already, but people are forgetting how insanely high the bar is now, LLMs do not even remotely begin to fix the problem of your competition being incredibly stiff.

Just look at something like ludum dare and all the top entries (out of thousands of games submitted) are all usually quite polished given the timespan.

As a Steam-based game developer, I am starting to use AI more and more in new projects for asset generation (images and text) as well as some help with code. Here are some of my ideas for why we haven't yet seen a huge increase in steam game releases due to AI:

1: Even with AI, it's a lot of work to make a full game. When most people think "I have a cool game idea", they're usually imagining something polished and non-trivial, possibly 3d. You could make a short text adventure in a few days with AI, or a very simplistic 2d game, but anything more ambitious (like 3d) is going to take a lot more effort.

2: Releasing on steam requires you to pay $100. I imagine this is a substantial deterrent for "3-day projects", unless you think it'll sell $100 worth.

3: There's more to game development than creating assets and writing code. The author of the article recreated an existing game, which sidesteps one of the most difficult parts of gamedev: design. Creating a compelling game is surprisingly difficult. Granted, you don't need a compelling game in order to release on steam, but I myself have made many prototypes over the years which I've abandoned because the idea just wasn't as interesting as I thought it would be.

4: I've made a few prototypes with AI assets, and one issue I frequently run into with image generation is: it still takes a fair amount of work to generate the same character in different poses, facial expressions, outfits, etc.

5: There is still considerable prejudice against using AI to make game assets. I think some people (myself included) are hesitant to release a game with lots of AI generated assets at the current moment, for fear of public backlash. Eventually that will calm down and it will become more socially acceptable to use AI to generate game assets.

I am bullish about AI improvement over the next decade, and I think we'll gradually see all of these issues resolve themselves as AI improves. But at the present moment, it's not quite as easy as the article makes it seem.

I've been "able" to make likely decent games for a number of years now. Code and 3D I could potentially have covered. That's no longer what scares me.

Having a good and semi unique idea, is a rare. If I had a great game mechanic idea, the rest would be trivial.

Say you do get a good game loop together that you feel will be successful. You will also now need to loop in art teams for artistic direction, music, character design, etc. A good game loop isnt enough, it needs to be presented in an equally interesting and unique way.

Finally, there is the risk. There is a massive time investment in making games, and you are catering to an audience that is not only accustomed to pirating but finds it morally righteous to steal your work. This is why app developers prefer to make iOS apps. The customers are accustomed to paying and have little interest in pirating.

post-launch and even before that, your job becomes paying and convincing streamers to play your game constantly in the HOPE people start to notice it.

All of this stress and work to hopefully just make an ok amount of money. I have so many excellent games in my steam library by indie devs that gave up after one or two very successful games. And I doubt it's because everything was going so well.

Delegating all the fun of making games to LLMs and leaving only the boring part for yourself puts you in the infinite procrastination mode.

Watching LLM generating the code doesn't help with producing the dopamine.

I'm not going to say this is all LLMs but Steam game releases have exploded.

https://steamdb.info/stats/releases/

About 50 games are released on Steam every day. How much higher are you suggesting this number should go?

This is like asking "where are all the very polished and professional LLM books."

This is a great analogy. There's a wider Fermi paradox here regarding business productivity. Where's the 10x economic output?

LLMs can't help much with assets?

I’m sorry but the difficult part of making games isn’t the coding, it is making something that is appealing and enjoyable to play. An LLM isn’t going to help with that at all. How is it going to know if something is fun? That’s the real work.

Also the idea that a dev who could making a game in 24 hour would create something professional and polished in 3 days is a joke. The answer to “where are all the games” is simple: LLMs don’t actually make a huge impact on making a real game.

If they're like me, they finish something and then sit on it for a variety of anxiety inducing reasons. I've got a state chart library I'm sitting on that I'm quite hesitant to share...

Code is not the hard part of making a game

What model did you use?