I get what this is saying, and I partly agree, but at the same time, I still think that open world video game npc dialogue is one of the few cases where “program that generates plausible sounding strings of text” could actually be mildly useful. Yes, if you generate everything procedurally, especially the actual story, the result is going to feel rather flat, but I think this misses any middle ground between having all dialogue dynamically generated and having it all be a prewritten human script. For example, suppose that an NPC had conventionally human-written dialogue for the most plot-important conversations, like giving a quest, cutscenes, a player’s first visit to a location, etc, just as these sorts of games have now, but during less important conversations or once the prewritten dialogue was exhausted, it would switch to generating new dialogue, using a human-written backstory and the prewritten dialogue for that character as part its prompt to keep every character from sounding exactly the same. Such a “shift” past what the writers had written for a character would be noticeable, sure, but so is just having that character repeat itself over and over, the utility wouldnt be to have AI write the story instead of the devs but instead to make the point in which that character’s part in the story is done less immersion breaking.
That sounds great until the NPCs start hallucinating.
Have you heard of the elf kingdom in the north? Legend has it that a weapon of unlimited power can be found there.
Then you search for hours for this “elf kingdom” only to find out it made that shit up.
Ah the irony of AI. Too unreliable to play an NPC in a video game, but reliable enough to pick bombing targets in real life.
Anything that removes the decision making… “I didn’t decide to bomb that girls school, the AI agent I asked did, so I’m guilt free.”
Ive had random people on the bus tell me random shit that was also likely untrue. NPCs halucinating random time wasting shit for players to waste their time on actually sounds more like real life.
The difference is real life has immeasurably more entropy and variance and randomness. Fiction is created, with intention, and good open world game design is one that leads and pushes and hints at things, preferably without being explicit - unless this explicitness is itself (mis)leading you somewhere or is just the McGuffin of the story.
Fiction without intention, even in its smaller, seemingly random encounters and details, is just messy and disrespects your time.
Regular reminder that all generated content is “hallucinated”, and saying the undesired output of an llm is akin to the program “thinking” wrong is just playing into the hyped up bs techbros are trying to sell.
No, in the case of llms this is not the appropriate use, but hallucinations in the context of “ai” is a specific phenomenon.
For instance, in automated transcription, hallucination is when noise is incorrectly interpreted to be coherent speech and entire pieces of conversation are steered in wildly different directions.
What? That sounds amazing. Especially because I’ve been thinking of the Insomniac Spider-Man games this whole time. That would be a great break from “‘Ey Spideeey!”
Ah yes, Canada. Famous for it’s fabled elf kingdom.
Who else do you know that harvests the lifeblood of ancient trees to refine into nectar?
I think of how many guards across Skyrim used to be an adventurer before they took an arrow to the knee, and think LLM output would be an improvement there.
Let the writers write the stories and quests. I am far less attached to what random background NPCs have to say, but it’s nice they’d say something.
I feel like this comment covers this pretty well: https://lemmy.world/comment/25336159
Every time I read something like this I realize just how little people understand how these things work.
The tiniest LLMs you can run on your machine are 1B, those take around 10TB of unrepeatable and well varied text to train. Let me put this into perspective, if you downloaded the whole Wikipedia, you would get 1% of the amount of data needed to train the tiniest LLM you can run. Do you think you can ethically source the remaining 99%?
What kind of game size should we expect then? Or have another “reason” to let a single-player game connect to the internet all the time so it can make API calls to remote datacenters? Isn’t it hot enough? Aren’t we in a bad enough point in time in this planet’s climate and economy?
Not to mention that having such NPCs would become the perfect excuse for the executives to expand the initiative, gradually onto the more regular and crucial NPCs and story bits, because to them, the more they can replace with AI, maybe the more KPI they get to meet and grow even fatter wallets, which is really all they care about - that will likely be coupled with increased pressure on the remaining writers, what with the beloved downsizing and all, rapidly leading to poorer quality. I don’t see how AI slop dialogue lines that definitely won’t be as instant as it is now, because whether it’s a local model or an API call, you’re subject to waiting for the lines to generate and make it to you.
There’s just no angle where even this case of only having AI take over backdrop NPCs is justifiable. The little fun that is poses is only appealing because of its novelty, but it won’t lead to a revolution in engagement and rich worldbuilding or interaction that would make the resources wasted make any sense.
It can be done much better and with magnitudes less of calculation:
- Write a number of phrases/phrase parts that can follow one another and make sense
- Randomly choose next one to be voiced by an NPC
- Profit
No lollygagging
Uh oh, something tells me the gaming community is going to prove them right.
There are already comments, even here, saying that it’s okay for filler NPCs and non-crucial interactions. And with the executives, this might be the perfect start, an Overton window that they slowly (if that) push open further and further, replacing more and more of the human in the art process in the name of profit.
The latter is especially disappointing, not because these people don’t care or understand art, but because the profit here is illusory, almost entirely upstream, lies in pure speculation, and kills the planet and economies at a pace that’s practically unseen. They only do it because they think everybody else is doing it and they can’t be left behind, lest the line doesn’t up and they don’t talk about the current trend on another bullshit meeting.
The corporate world is such a mistake.
This machine has been killing everyone who can’t work its gears within the top 10% of workers, for far far longer than the AI era. The problem is that now most of the former cog-turners are watching this truth unfold upon them instead of upon others.
The leopard always eats the face.
G*mers made a big deal out of boycotting Sony.
Only to turn around and celebrate Palworld.
G*mers are dumb fucks
I think there is a very narrow, specific implementation of LLMs for NPCs that could be used for games that I personally would deem as both ethical and acceptable:
The model would need to be trained exclusively on a deep write-up of source materials authored by human beings (something like what Patrick Stewart was given for his role in Oblivion); and human voice actors hired for all characters, who explicitly approve their use towards generative characterisation solely for this one role/game.
This is just my personal line in the sand, and I know not everyone would agree - and that’s fair enough.
Every time I read something like this I realize just how little people understand how these things work.
The tiniest LLMs you can run on your machine are 1B, those take around 10TB of unrepeatable and well varied text to train. Let me put this into perspective, if you downloaded the whole Wikipedia, you would get 1% of the amount of data needed to train the tiniest LLM you can run. Do you think you can ethically source the remaining 99%?
10 Terabytes. 1 letter is 2 bytes in unicode. 10e12 bytes. 5e12 letters. 1 word is about 5 letters. 1e12 words. Assuming 500 words per page, that would be 2e9 pages. Page thickness assumed at 1e-4. 2e5 meters, if you stack all the pages. 200 kilometers high tower of paper.
<pedantry>For English language text letters are 1 byte when encoded in UTF-8 (which is how essentially all text is stored and transmitted these days).</pedantry>
Though it probably balances out in your final answer since you’re not counting spaces, punctuation, formatting, or other markup.
Meaning that to train even the tiniest LLM you can get you need a stack of paper so tall it could get knocked by satellites.
This is the scale that people miss when they talk about “writing some stuff to train an AI model”.
I don’t think they care about that, it’s just about profit. They’re making rich people richer (or they’re out of a job) they’re not making art.
It’s a bit late for that, the gaming community seems to be fine with bad stories and mechanics in modern mainstream games.
studies show that people prefer stories written by AI over stories written by humans
the author states his “general distaste for generative AI” but that is not supported by the science
Show some fucking proof
These people really are fucking delusional.
Im sure that the studies proved that entirely derivitive work will always outstrip origional ideas.
/s < edit, added for clarityLet’s do a study right here and now using your upvote/downvote ratio and see if that holds up.
