The Elephant Shitting in the Room
Yup, it's the LLM post. I didn't know what to write, but considering the discourse, I suppose I should get this one out the way early.
Chances are, if you are remotely interested in games, you have an opinion on LLMs, GenAI, whatever the grifters of Silicon Valley decide to call it this month, and considering the people around me there is a good chance this opinion is negative. From the energy needs (leading to environmental damage and water scarcity), the blatant scraping of everything on the internet without consent, to the societal damage, there are plenty reasons to hate on the technology wholesale.
This isn't about that.
This is about video games.
Disclosures Aren't Enough
If you've seen me as a reply guy on Bluesky, you probably know that I'm staunchly pessimistic when it comes to AI Use Disclosures. This isn't because I'm against the idea of LLM use being disclosed, but because these disclosures simply never go far enough.
The one that people usually point to is Valve's AI disclosure questionnaire, which reads as follows:
"Pre-Generated: Any kind of content (art/code/sound/etc) created with the help of AI tools during development. Under the Steam Distribution Agreement, you promise Valve that your game will not include illegal or infringing content, and that your game will be consistent with your marketing materials. In our pre-release review, we will evaluate the output of AI generated content in your game the same way we evaluate all non-AI content - including a check that your game meets those promises.
Live-Generated: Any kind of content created with the help of AI tools while the game is running. In addition to following the same rules as Pre-Generated AI content, this comes with an additional requirement: in the Content Survey, you'll need to tell us what kind of guardrails you're putting on your AI to ensure it's not generating illegal content." AI Content on Steam
Note how it says "code" in there - how many AI disclosures on Steam have you seen that mention code? Very few, if any. However, the problem isn't even that they would likely not bother to check if you have traces of Claude Code (and there are almost always traces).
It's that they only check what is in the game.
If you use a proprietary tool created by a person using a LLM, that doesn't count (Note: In this case, I limit it to proprietary tool as otherwise any 3rd party engine would likely get flagged as they practically all have a LLM component somewhere even if you don't use it). If you generate concept art, you can skate by because unless you tell on yourself, you don't have to disclose it.
Worse, there doesn't seem to be any consequence for not disclosing and getting caught/telling on yourself. An example: Blood of the Dawnwalker used a LLM to generate voices for testing purposes with no intentions of leaving them in the final game (and for all intents and purposes, they didn't as this is information they volunteered themselves). The rest of the game, according to what I've read, does not have traces of LLM use. It did not disclose this, and are currently facing zero consequences.
The only consequences are social at best, and when you're a Polish (or anywhere else) studio who will likely never really see the backlash against LLM use... do you care?
You might be thinking that I'm calling for us to abandon the fight. Absolutely not. Even if it was the most ethically sourced dataset of all time, even if it was environmentally safe, even if it wouldn't have the negative social effects we're currently seeing, I still would be against it. Why?
Cut So Many Corners, You Are Now The Orb I Am Pondering
I am, unfortunately, in a situation where LLM use is present where I work, and every single person who has continually touched the damn thing has become worse at their job in some aspects, because the very essence of this technology is not to make things easier, it's to cut corners.
When you use LLMs, you're not actually making the thing. You are telling something else to make the thing for you. You effectively become a manager. How many managers have you seen actually check if the product functions? I've seen them but it was more to find someone to blame than it was to actually make sure the job was well done.
As to how LLM use leads to cut corners? Most of the examples I have would be proprietary, but I'll try to give up a few. The first is simple: Even if you are able to verify the code, you likely won't do it. If you couldn't muster the will to code the thing, why would you verify that it can be presented for either design or QA? If you can't verify the actual code, why would you do a smoke test to verify if it works if you're not the person who'll actually use the thing? These are basic things to ensure that nobody else wastes their time and that you didn't waste yours. However, if you have the LLM handle it, you won't even take that into consideration as that would be admitting the LLM can be faulty and make the entire operation slower.
Every cut corner leads to additional laziness from the person that cuts corners, eventually leading them to become a liability on the team because they simply don't work, just tell something to work. If they ever need to actually work, they'll find that they can't.
There is a quote from Arcanum: Of Steamworks And Magick Obscura that applies, doubly so as it is said about the man who stole the steam engine from the dwarves and distributed it to humanity - Gilbert Bates. Troika was a lot of things and subtle wasn't one of them.
When Bates was given a look at our technology, he was overcome. As a human, his first thought was 'What can I use this for?', when it should have been 'what is the cost of its use?'"
Competence is the cost of using LLMs. If the LLM actually was a force multiplier like they claimed, that wouldn't be the case as you would still need knowledge to make something and would not lose that knowledge the more you use them. I've seen developers forget the actual workflow and forego documentation, making it difficult for QA to know if the data that was entered was correct. I've seen developers admit they didn't even read General Design Documents anymore and just had a LLM summarize them, making the end result just as vibe coded as if they had used the LLM to make the feature outright.
My fear isn't that LLMs will take jobs (ask translators and writers, it already has), but that people who used to be competent that are forced to use these things will find themselves digested into elephant shit when the bubble inevitably pops while the rest of us in games will be covered in pachyderm doo-doo.