For PC gaming news and discussion. PCGamingWiki
Rules:
Been saying it since way before chatbots sent everybody off the deep end, it's not about the holodeck model where generative AI makes things for you out of whole cloth, it's about letting the same artists that are already doing this well do it faster.
A lot of the hostility comes from being misinformed about the tech, but the advocates aren't doing any favors by also completely misunderstanding the tech.
letting the same artists that are already doing this well do it faster.
Not just doing it faster, but i think there might be some benefit to allowing AI to generate content that would NEVER have been made otherwise.
Like inane NPC chatter. We could have virtually unlimited NPC conversations, instead of "i took an arrow in the knee" over and over again. Would that make a game better? I dont know. Maybe, maybe not. But what i do know is that endless NPC chatter is not something ANY developer would EVER take the time to do so why not let AI do it (if it makes a game world and lore better)?
I would love aigen chatter. I like the idea of free talk, but I am also concerned because NPCs can give away info early or even bad info.
There is an absolutely ludicrous amount of NPC chatter in many games out there today. And specifically the writing of it is definitely not the problem. Any competent game writer can deploy reams of the stuff at the drop of a hat, way more consistently and effectively than ML generation.
Plus you still need to edit it, plug it in the right places and make sure it's driven by the right logic. Any asset in a game requires work, regardless of what parts are automated. It's not like your average GTA city feels empty and silent. If anything there's too much of the stuff sometimes.
But if the time and effort of doing all that work can be cut in half, or by 90%, then you can have the best of both worlds: the hand-crafted, edited, art-directed stuff that fits where it needs to fit and reinforces the artistry of the game and the breadth of content from being able to plug it into the right places without a whole lot of manual labor.
Procedurally generated content, particularly in filler areas, is nothing new in games. Since you bring up Skyrim, Daggerfall had tons of procedurally generated assets, dialogue, dungeons and areas all the way back in 1992 already. It's about making those better by either allowing for humans to supervise a larger percentage of the filler or by making the filler of slightly higher quality. If I'm excited about generative AI applications for creativity at all, I'm excited about those things.
There is an absolutely ludicrous amount of NPC chatter in many games out there today. And specifically the writing of it is definitely not the problem. Any competent game writer can deploy reams of the stuff at the drop of a hat, way more consistently and effectively than ML generation.
Yet.. they don't? If it's so easy, why do we constantly hear the same chatter in major games like Cyberpunk? I don't think your definition of "ludicrous amount" applies to games people put hundreds of hours into playing.
It absolutely does, although it's fine to have different expectations for what we want, I guess. Cyberpunk's open world design in general has issues that aren't down to modern AAA studios' capabilities to generate assets, so I don't know that it'd be my gold standard, but you're getting 200+ hr games these days where pretty much every NPC is voiced, which is absurd.
Now, in terms of priorities, would I download a 300GB game for the sake of a bunch of AI-generated sound files to ensure every incidental conversation is unique at the additional cost of having the filler chatter sound worse than fully voice acted lines? Eh... probably not. But I'd love it if the devs were able to have more variety without being soul crushed by having to do repetitive work around it. And I'd certainly love to see a higher percentage of the same 100-150GB game brought up to "hero" quality instead. Your mileage may vary.
although it’s fine to have different expectations for what we want, I guess
I don't think having the expectation of "every time I go to a place, the same two NPCs are having the same argument outside" is wildly absurd. Reminds me of Black Desert, one of the major banks had a small child berating a large man non-stop, EVERY DAMN TIME you had to stand at the bank managing your items. It was intolerable.
Alright, but you do realize that specific examples of NPC chatter being poorly implemented doesn't immediately lead to "AI writing and performing NPC chatter", right?
Those examples stand out because they're bad. The arrow to the knee was a tuning issue, not a lack of assets. Getting excessively noticeable repeated chatter in a frequent revisit location is a design issue.
The idea that the frankly atrocious example in the video is the solution to that, instead of the tons of examples of actually good NPC chatter that you near all the time and don't notice is a bit of a leap.
Was this completely AI generated? The graphics weren't even bad. The delivery of the lines might be one of the worst things I've heard and seen since Jesus Christ Vampire Hunter, but this article has a clear bias against AI. I get it, use AI to iterate logic and mechanics, let the fucken people write and talk. But that doesn't mean this demo is 100% steaming pile because AI was involved.
Jesus Christ Vampire Hunter was high art, don't try to front.
Jesus Christ Vampire Hunter
I have never met anyone else online who has seen this movie. Thank you for consuming Canadian Art.
Not the graphics or gameplay... But the creative writing.
Watch the trailer. The NPCs, the dialog... Ooof
Oof indeed. I mean someone has to work at it if its going to get good. But having AI do anything except be my tool as a programmer feels taboo. Especially if we're going to continue doing this capitalism thing where people actually suffer because of this.
I can't believe anyone would look at that demo and think anything remotely close to "I'd play that for more than 2 minutes".
This isn't for players. This is for developers. A proof of concept does not mean its fun to play. It means it technically is possible to do, how do we make it not suck?
I for one grew up in an era where voicelines were "Huh, huh! Huh, huh, huuh!" Just marvel at how far we've come in barely 30 years.
The video they released is abound with some of the flattest readings I’ve ever heard. The player character at one point blandly demands to know “what the gangs are up to.” The next scene has them ask my favorite line: “Hey, what is that tree over there?” A robot, which ironically has the most spirited delivery in the whole trailer, flatly responds, “The tree is the hologram of Yggdrasil, the Norse realm tree.”
While it shit writing for a big budget AAA narrative game like Cyberpunk or Witcher... It's peak writing for something like a JRPG.
Intonation in virtual voices will get better. The voices that OpenAI use have good conversational intonation. If this company can make or use good models isn’t clear though.
Apart from that, I’ll be surprised if any of this procedural stuff can actually turn out interesting. Even with reasonable synthesized voice acting.
The delivery of these lines makes them feel like the ones you see in unfinished content where they have some random person do the lines because they haven't cast a voice actor for the part yet.
Who would've thought using tech that rips people off would just always suck lol