https://nitter.1d4.us/TheUnaButters/status/1651778740421287941
This sounds genuinely wonderful. Endless flavor text would be a boon for making the world feel big, y'know?
I genuinely like the idea of AI subquests and flavor littered through a digital world, rather than a set of canned "Beginner Quests" that you just look up a walkthrough on in order to breeze through.
But this still just feels like shit for people with too much time and not enough friends. More opium for the masses, so you don't feel so fucking lonely all the time. It just feels like its going to be the thing that leads to "I married an NPC in World of Wankers, but now the server is shutting down and I want to kill myself" posts.
AI subquests and flavor
'radiant' generation already makes up a large part of skyrim & later games' content. AI is like a hat on a hat.
you want depth, you've gotta do it by hand. you want shallow, keep it simple and use the tools we already have.
a hat on a hat
I am once again asking you to stop giving Valve ideas
you want depth, you’ve gotta do it by hand.
At some level, the computer is going to have to handle everything. Otherwise, you're just playing table top.
But in the same way you can marginally improve graphics, you can programmatically improve story.
That does require craftsmanship. It can't just be throwing prompts at ChatGPT lazy. But you can outsource things like fluid dialogue and ambient chatter to an AI to flesh out the background of a game in a real way.
Enough so that its not just a soldier doing a hallway patrol in a loop at all hours. Or a monster popping into existence and then getting baited into an easy ambush over and over and over again.
dialogue is the only thing we're talking about here, procedural generation and game AI sophistification are increasing through labor of human devs and hardware capabilities.
and i'm seeing extremely limited utility for AI text generation if you're gonna be approving all the lines & handing them off to voice actors anyway. we dont have a million chatter lines because they aint paying a writer to spend time on it, which they still would in that case
approving all the lines & handing them off to voice actors anyway.
Nah, you'd use a voice synthesizer to do it on the fly. That'd be the whole point of it, being able to come up with novel responses based on teh question asked.
You need to remember that these LLM things aren't smart, they don't think, they aren't capable of reflection or semantic understanding. All it's doing is generating strings of text based on weighted probabilities. That's fine as long as you keep things very superficial, but if you try to get any depth out of it the gaps are going to become glaringly obvious very quickly.
You know how the "art" models can't draw hands? If you try to use these things for more than they're capable of you're going to get that kind of unnerving uncanny valley fuck up in your narrative. They can't draw hands because they're not capable of abstraction. They don't have an abstract understanding that a 'hand' is an object with certain characteristics no matter what position it's in. It can't do hands because they appear in so many different shapes in so many different contexts that it can't identify a consistent enough pattern to reproduce the hand. Faces are relatively consistent and easy, but anything that moves around too much - hair, hands, really almost any detailed object that's not a face, it can only draw very superficial approximations of. Jewelry is a good tell because there are so often pieces that don't connect to anything, or pieces that flow in to flesh or clothing.
You know how the “art” models can’t draw hands?
The latest version of Midjourney is now able to draw hands with a fat higher consistency and quality. It only took them a year to polish that bit.
It can’t do hands because they appear in so many different shapes in so many different contexts that it can’t identify a consistent enough pattern to reproduce the hand.
There are a lot of simple end runs around this glitch, even before the upgrades and refinements. Just requesting hands appear under an apron or behind the back, for instance.
Professional artists do shit like this all the time. There's a whole rant online about how Rob Liefeld can't draw feet. He's been churning out comics the old fashioned way (by stenciling over images he's not great at drawing) for decades.
All that is to say, we're not asking AI to work miracles here. Just to improve over what is rudimentary and generic digital conversations in a way programs like Replika have already proven successful at.
Yes, if you start too hard at the edges, you'll lose the illusion. But we're not trying to outwit Harrison Ford in Blade Runner, here. Humans are very easy to fool when they want to be fooled.
opium for the masses, so you don’t feel so fucking lonely all the time.
About time. I don’t want to watch marvel.
Endless flavor text
this would so easily compromise worldbuilding and plot unless you've just got an AI imagining 10000000 ways to talk about the weather
If you ask chat gpt to answer things as a Marxist leninist then it keeps things in the ml universe so to speak. I don't see why you couldn't ask "as an elder scrolls character" or "as a dunmer bard" or whatever and get something that's mostly in-universe. TES lore contradicts itself so much just from arena and daggerfall to morrowind that I don't think it could be any worse than the in-universe contradictions.
specific to Elder Scrolls, you've got a bit of a point with inconsistencies but even if you got it to remember the things it generated in a playthrough or instance, it'd be maddening once you try to take any information and apply it to anything else.
a dunmer bard telling me the weather in Blacklight is ashy, rainy, snowy, or sunny in separate playthroughs. like at that point i don't care to know. what's the point of learning additional 'lore' if it isn't actually lore?
Hmm, it would take a lot of work shopping. I still think it's a powerful tool and there's definitely an application for open world RPGs
I think it could work if you went the other way around. Like you have a huge database of facts about the world, and then the chat algorithm is capable of spitting them out or working them into conversations naturally. You could make an entire game out of this - think of the game Her Story, where you have to search through the interrogation clips to find the information you need, but instead the information you need to solve puzzles in the game world is in the chat algorithm, and you need to talk to the right NPCs and ask the right questions to find it.
Yeah but thinking people tried to tie the elder scrolls lore in to something semi-coherent, generating a lot of the most interesting concepts in the setting along the way. A language model just spits out strings of words with no semantic meaning. The LLMs can spit out marxist sounding gibberish because they're re-mixing the contents of untold millions of lines of text scraped from the internet. The LLM has a vast amount of information from which to cobble together a mockery of human speech.
For an elder scrolls game you're not going to have that. you're going to have a few hundred pages from the books, and whatever additional information the writing team comes up with. It's going to be far, far more restrictive. The LLM can't make things up, it can only repeat variations of what's in it's model. It has no self awareness, introspection or reflection so at most they'd be able to hack together a very limited way to keep it's responses consistent across multiple NPCs.
These things are very limited. I know people are having fun reading the text they spit out, but the number of people who still think the LLMs can determine the truth value of the statements it generates or "know" what its talking about is frightening.
Sounds good and all, until every NPC is shouting
"A NEW HAND TOUCHES THE BEACON"
HEAR ME AND OBEY! A FOUL DARKNESS HAS SEEPED INTO MY TEMPLE! A DARKNESS THAT YOU WILL DESTROY!
random npc walking between the conversation
HAVE YOU HEARD OF THE HIGH ELVES
Every NPC proceeds to go on their own personal diatribe about mudcrabs
Walk into Solitude, the Nords start saying really uncomfortably fash shit but since it's about dark elves the nazi filter doesn't catch it
That would be kind of cool in a horror game. Having all the NPCs start shouting your own words back to you in your own voice would be creepy af.
I have a feeling it would start to feel like noise pretty quickly. We already have games that use procedural generation extensively to create endless "variety", but the patterns become noticeable almost immediately (every planet in No Mans Sky, for example). Not to say I would mind if every NPC said something slightly different instead of hearing the same 10 pre-recorded lines, but it feels like this wouldn't have much impact.
I guess I could see the benefit if the NPCs changed their dialogue based on the player's actions in the game. It would be fun to hear townsfolk talking about things that I was doing, or events that I was involved with or directly caused. It would still be the same kind of noise, but it would be a fun way of making the player feel like part of the world.
I think the real test here is whether or not you can make the interactions of patrons in a tavern actually worth spending 5 hours in like you would a pub in real life.
If people can have real genuine fun at the bar shooting the shit with the patrons and playing bar games then you could actually add a layer of depth and immersion to open world games. You would need to change the entire pacing of the games of course to something more like Kingdom Come, but it could work if you implement it right.
I don't have much faith in the techbros knowing how human interactions should look and feel and how real people with emotions enjoy their time with one another though.
The pacing issue is the major one here however. As you increase immersion in these games you also must decrease the pacing down to an actual real life pacing. This is an issue when it comes to travel however, which is less fun.
I can get a paragraph or two into talking with ChatGPT before I start to get bored of how often it’s just rephrasing what I’ve already said. It was more fun without the guard rails that turn every other message into a disclaimer.
It was more fun without the guard rails that turn every other message into a disclaimer.
:100-com: completely agree. But the technology doesn't have to be chatgpt. Others without its limitations will appear. I think the kind of thing I'm talking about could emerge.
worth spending 5 hours in like you would a pub in real life.
That's asking way too much. I think the farthest you can, and should, go with this is having idle background chatter that's being generated. Just simple, low risk stuff like talking about the weather, or mentioning significant recent events based on keywords.
These language models aren't people. It doesn't think. Trying to make it simulate an entire bar full of people for hours and have it pass the turing test isn't going to work, but you might be able to get it to do simple conversations between NPCs, and contextually appropriate generated answers to natural language questions from the player.
But it'd be more "How do I get to X landmark" or "What happened to Y NPC?" than having detailed conversations for hours.
If you used it judiciously and were realistic about it's limits I think it could be cool. Being able to ask random NPCs "How do I get to Castle Doom" and get novel responses in natural language each time could be cool. Using it as a way to inject backstory without having the character read tons of books could be cool. Using it to make NPCs have novel background chatter could be cool. It doesn't have to be flawless, just good enough, and you can, or at least should, significantly limit the scope of it's responses. Have your writers build a big document with information the NPCs should "know" and get the language model to work from that.
Absolutely. A lot of those suggestions I would consider to be over and above merely "endless flavor text", which to me implied taking more or less the current way of doing flavor text and throwing AI at it. I think you still have to structure it and design the parts into a coherent experience, or else it'll be perceived more or less the same way as static flavor text. I don't think that would necessarily be a bad thing, I just think we'd get used to it pretty quickly.
:geordi-no: Endless flavor text
:geordi-yes: Endless Flavor Town :guy-fieri-shining:
More importantly, Boimler is using the holodeck programme as an oracle/therapist which is what people are using ChatGPT for right now.
it truly is exhausting being a morrowind fan and having all the oxygen in the room sucked out when skyrim gets mentioned. get better taste, you're being bad nerds
i've been playing OpenMW. alchemy is so OP lol. infinitely stacking potion effects. I just brewed and drank intelligence potions until each potion was worth like 50k
In fairness, quite a few Skyrim fans also seem to be Morrowind fans. And some of the best content I've ever played since Morrowind has been from Skyrim modders who want to inject some of that "magic" (Enderal, Beyond Reach, Legacy of the Dragonborn, several others).
Then again, most players supposedly don't even play with mods so :shrug-outta-hecks:
I remember an indie game that did this, with a couple who were arguing and would decide whether or not to get a divorce based on your actions. It's a neat concept but far from shippable, the Skyrim mod in the OP basically just has the NPCs reciting facts about themselves which is about what I'd expect. The campfire conversation where Lydia just gives you a wikipedia summary of her own personality is :data-laughing:
To really make this work, you would need a small game world with maybe a dozen or so NPCs, loooooong backstories written in painstaking detail for the speech generator to work with, and a finely-tuned generation algorithm that doesn't just pick relevant bits out of that background information and spit it out at you but that actually iterates on it.
I love Facade. You can get the door silently slammed in your face the moment it's opened by just saying, "Maria's melons."
Ten years ago I actually made a YouTube video where I got the door slammed on me by asking if they wanted to hear about Jehovah. I also saw someone else do the phrase "So I heard you're getting a divorce?" which has stuck with me all this time.
My taste has changed so much. Back then I subtitled that video "I hate pretentious games like this", but nowadays half of what I play is probably something my 20 year old self would call that.
Façade creates the illusion of Trip and Grace taking to you by making them really unpleasant people, irony of ironies. They brush you off if you say something not in their scripts, often changing the subject to whatever the next event is supposed to be, and aggressively push you into yes or no ultimatums all the time. Naturally you get the ending where they don't get divorced by exploring the apartment and finding topics which weren't part of the perfect evening they planned.
