Hacker News Reader

Steam's basically already flooded with low quality games, but the recommendation systems (human and otherwise) seem to do a decent job at dealing with it.

Steam already has issues with fake reviews and other scams. I doubt they'll manage to clean them up any better when there's a sudden flood of garbage to take down and within minutes after a bad actor gets banned they can create a new account and publish a new "game"

True. Perhaps it will end up not quite as bad as it could be for the lowly consumer, but for game developers who do put in the effort instead of focusing on store visibility tricks it will inevitably get even worse than it already is. Remember when the YouTube suggestion bar wasn't dominated by impossible photo edits, cleavage and/or that shocked face? Me neither.

YouTube's fine for me mostly, unless logged out which is a nightmare.

I think that games are a medium where it's a lot harder to flood a store with garbage, since you've got to go through and actually make the game. There are plenty of shovelware games on Steam already but they're using marketplace content that's been created by real humans for game systems. If you're going to make some shovelware garbage you've got to put a lot of work in - game systems, saving, loading, achievements - and every game costs you the Steam creator fee. Also, 3D model generation isn't up to scratch yet, so it's only going to work on 2D elements.

My game has a few elements which I used AI for, but I don't think that it's particularly obvious which. Here's a screenshot: https://drive.google.com/file/d/1yM_3Ih40breLznsqy7wrA0y7_iF...

> I think that games are a medium where it's a lot harder to flood a store with garbage, since you've got to go through and actually make the game.

That was only true before you could tell an AI to make a game for you while having zero programing skills, zero artistic skills, and zero writing skills. The better AI game generators get at creating whole games from nothing more than descriptions the worse things will get.

You couldn't make an entire game without those skills. In 10-20 years? Maybe. But at the moment unless you know how to code you'll have a million compiler errors and no game as the LLM hallucinates half the engine functions and gives you a ton of unconnected code which you can't put together. It's the same with the artistic skills, unless you know some artistic basics you're going to end up with art that clashes so badly (different styles, different themes etc) that people will be put off just by a screenshot.

You could use AI to make entire games without those skills at least three years ago!

Here's one from 2022 https://andrewmayne.com/2022/03/17/building-games-and-apps-e...

Here's another: https://nitter.cz/ammaar/status/1635754631228952576?lang=en

sites like this show where things are heading: https://www.rosebud.ai/ai-game-creator

I'm not too impressed with these examples. I mean it's impressive that it can build code, but it's a very long way off building entire games that you'd want to play.

The human still has to provide the creative input - the concept of the game, the rules, and multiple steps of guidance to get anywhere (when it works).

The code is sometimes generated correctly, but only if a) it's fairly trivial and b) the model can generalise from something else it's seen. There's squillions of maze examples out there so models build slight variations on this easily enough, but as soon as you inject some originality, the model flounders.

Try something more recent:

https://www.techspot.com/news/100718-someone-created-video-g...

Play it yourself here: https://bestaiprompts.art/angry-pumpkins/index.html

This is exactly the kind of crap I expect to see on steam

He had to know how to code to tell it what to fix and diagnose what the errors were and it sounds like it was more arduous than simply writing it himself, so while technically it was entirely written by ChatGPT it still seems like it's not going to make a big difference in the short term - it wasn't just "ChatGPT, write me an Angry Birds clone".

"I think that games are a medium where it's a lot harder to flood a store with garbage, since you've got to go through and actually make the game."

Have you not seen mobile app stores?

I admire your optimism. Look at the iOS App Store to see heaps and heaps of trash games. Even without any “AI” support.

Not really. See: https://arstechnica.com/gaming/2019/03/a-dev-trained-robots-...

In this article a developer made a completely automatic game generator for Android. A terrible slot machine game with graphics pulled from an image search, generated names, sound is text-to-speech, etc... Even the application process was automated. In the end, they published 1500 apps at a rate of 15/day, the maximum allowed by the Play Store. They are now all dead, but in the meantime, they made $50k.

Would this be any worse with generative AI though? You'd just be replacing the images from Google with images from Stable Diffusion. You couldn't let AI generate the code for it because you'd have to actually check that it successfully compiles.

Hmm this sound like a bad deal for that developer. Looks more like a performance art than business plan. The amount of work needed to automate all that stuff probably took more than a few months for only 50k income.

> steam is going to be flooded with massive amounts of games that are nothing more than AI generated trash.

How is this different than the floods of human generated trash that shows up from time?

The volume, I'd presume.

Massively higher potential output. It's the difference between sending spam e-mails out by hand and having a bot do it.

Output will not be much higher compared to the Unity store's asset flips. AI still gives you less types of content compared to Unity store (which provides you with whole game mechanics/features/scenarios that you can use). AI will lower barrier a little more - you will not need to buy a lot of resource packs anymore to finish a game.

The only thing that will change is that it will be little harder to detect if the game is just asset flip or something more.

This still translates into being a worse state of affairs for the store, and for the consumers who now have to contend with more scams that are harder to detect before pressing "buy".

I never had problem on Steam to find good games. Steam also very rarely recommends something really bad.

Not sure how it is implemented but for now crappy games are not even visible on the main page (even on the new games list or soon to release list), you need to expand those sections to really see every game.

Steam's refund policy makes this a non-issue. It takes way less than 2 hours to figure out you have a bad game.

It certainly helps a lot, but I wouldn't call it a non-issue. It deteriorates the user experience. From the scammer's point of view, with a low barrier to entry, they don't need very many people to fall for it before some number of them will fail to get a refund for whatever reason.

Even if literally nobody loses money, it's adding noise to signal.

Steam is already filled with low quality games, without any AI. But they're mostly either unremarkable clones or one-off shitposts made as a reaction to something, not "low-effort trash to get rich quick". This scheme doesn't make much sense on Steam which is organized like a ladder where games compete for attention, unlike the books on Amazon which is flat in structure, and your low-effort garbage can pass for a legitimate book at first and is guaranteed to attract someone.

Cheap AI-generated features will fill every game, of course. It's inevitable and will soon pass, just like every trend - bloom shaders, piss filters, cloning minecraft and battle royale, etc. Some of it will stay, like hiding the lack of developer-hours behind the low-res pixelart. Some of it will be even good!

Steam is basically a monopoly, thus it's good that they allow the maximum amount of games to be published on a free speech based policy, otherwise they would have justification to block god knows how many small indie titles on bogous claims and the axe of the legislator would surely come their way

Well I for one am waiting for someone to do proper LLM integration with some dialogue driven game, as unreliable as it currently is. How much more hilarious would Fallout be with the option to just say anything to anyone with genuinely matching in game consequences instead of picking one of four lame options.

For that to be actually good, you don't just need AGI, you need an LLM with the writing quality of a New Vegas writer. The difficulty there is that the training corpus is too small to create one.

Eh not really, you just need a good way to RAG the existing story so it only mildly adapts it to your actions. No need to make up the whole thing on the fly, baby steps.

A prompt along the lines of character card, available actions, story so far, what player said, response text, response action. Actions still limited to what the game can actually do of course, but it would already be far more varied than strictly defined trees. Shame that local llms are so crap at function calling for the actions part, but it might stil work with an ad hoc model and extensive grammar restrictions.

No. I think people will come to understand and accept that what you get out is commensurate to what you put in. If you stick close to the script (there will still be a scripted main story in such games, because people want shared experiences) and say and do reasonable things in the setting, the LLM game master should be able to deliver reasonably, and improvise around the edges. But if you go wildly off the rails, then no LLM is going to turn that into an award-winning story.

"How much more hilarious would Fallout be with the option to just say anything to anyone instead of picking one of four lame options."

Only if those dialogues lead to something real. Just doing small talk with NPCs is easy to accomplish, but integrating it into the world, where depending on the dialoge you get new quest relevant information - or have to fight the NPC if you insult them too much or say the wrong thing - is way harder.

True yeah, it may make sense to have multiple models or at least different processing stages for specific parts of that. One to actually reply back based on context, another to do sentiment analysis for facial animations, another to pick from a list of reponse actions based on the convo, another to adjust the storyline to make it make sense given actions so far, etc.

AI has the potential to be able to do all kinds of cool things. Currently, I wouldn't trust it for anything important. The more story driven a game is, the less you'd want to leave what currently passes for "AI" in charge. It'd be no fun if you couldn't complete a quest or the story because an NPC is depending on a hallucinating AI. For a little while it might be fun to convince a confused AI that you've got an item you don't actually have, or trick it into thinking that you're a developer in the middle debugging the game so it should just spawn whatever item you ask for, but I think that sort of thing will get old quick.

Check Vaudeville

I think this is a pretty drastic take. Shitty products are not the problem of marketplaces, fakes are. Fakes seem hardly a problem in the market for digitally sold games. Steam solves that problem quite well (Amazon as a counterpoint). I don’t care about the 12k games published in 2023, I bought a few after reading reviews. The risk of the 12k (with say 100 gems) becoming 240k (with say 102 gems) is pretty low, even if we allow for more false reviews on Steam becoming a factor. Steam could happily start listing MetaCritic reviews and other external reviews as additional sources and then there is again no problem unless I am missing a part of your argument. (And then we still have the service Steam can deliver counteracting this risk by refunding purchases, which they do after filling in some forms in my N=1 experience.)