Support NeoGAF

[Men_in_Boxes]

I'm late to the party but I just figured out how interesting / powerful AI dialogue from ChatGPT can be. IMO, the technology is already there to make characters far more interesting than what the industry currently produces.

Are developers fast tracking this technology into upcoming games or is there something preventing this from being implemented? Single player gamers of NeoGAF, educate me.

 

[ReBurn]

Next week sometime I think

 

[amigastar]

It could be nice to have the NPCs say you name you called yourself with AI generated voice just for the name woven into the npcs dialogue.

 

[AV]

They're already implementing it as we speak.

For a better answer, it'll never ever fully replace written dialogue for numerous reasons, and it's rarely ever going to be more "interesting" because LLMs are inherently based on what a human being has already written. But it's absolutely going to be the future of pumping out content faster and cheaper than before. The brakes are off, and nobody really has hold of the wheel.

 

[feynoob]

10 years for proper AI sound. Currently AI is dogshit

 

[simpatico]

We can only hope so. No real need to pay a unionized voice acting gang to provide small talk. Just another hurdle in front of small dev teams. AI is the only hope gaming has.

 

[Fafalada]

For your average game - models need to be compact enough first to ship 'with' the games - noone is going to burn money at AI service effigy for the sake of it.
Right now I think you need something in order of 10GB for a decent conversational model - and that's just for text (it won't be anything that competes with chatbots online - but it should be able to service some NPCs), no idea what the commercial implications of it are (I doubt most devs will have the time and $ to train their own for awhile yet).

So if the question is when will this show up in your JRPGs and VNs - probably a good long way off. We may actually see AAA being at the innovation forefront here - for a change.

 

[sendit]

10 years for proper AI sound. Currently AI is dogshit

You always seem to have awesome hot takes:

 

[T4keD0wN]

In reality its already used a bit in experimental projects, but itll take years for it to get good and used in proper games.

I hope it doesnt take off anytime soon, running it locally is a waste of resources and space that could be better used elsewhere and we dont need more always online games.

Funny answer would be that Ubisoft will be the first major publisher to heavily use it and nobody will even notice it.

 

[MarkMe2525]

Already exists, but itll take years for it to get good and used in proper games.

I hope it doesnt take off anytime soon, running it locally will require too many resources and space. We dont need more always online games.

I bet Ubisoft will be the first major publisher to heavily use it and nobody will even notice it.

I imagine they will refrain from running it locally. It will be the new DRM and they will just say they are trying to give us an immersive product.

 

[Loomy]

What makes you think it hasn't happened already?

 

[Three]

Speech synthesis in games, what's happened to it?

Are there any games with it? I know it still sounds somewhat robotic so probably doesn't make sense for most games but considering a lot of games actually have robots in them why don't we have this? Portal, Horizon forbidden west for AI, and Detriot for example. Portal would have been ideal...

www.neogaf.com

 

[March Climber]

 

[willothedog]

Cyberdyne Systems Nvidia are working on it.

 

[mrqs]

Until the tech gets into place and consoles fully support a good enough language model, it might take 5 years + to start seeing that in some small games. AAA games? I'd say 10 years because development takes too much time, and decisions like that need to be made in the concept stage.

 

[feynoob]

You always seem to have awesome hot takes:

So 10 years then. Because this is still early stuff.

 

[DaciaJC]

I think it will have fantastic utility in mods, where passion-driven people working for free obviously can't afford to spend much money on voice actors. Exciting prospects!

 

[feynoob]

Cyberdyne Systems Nvidia are working on it.

Soo robotic. They will need to make it real. Was having a hard time listening to them.

 

[Pejo]

If it ever does, I think that will likely mean permanent "always online" single player games. If you want the best quality AI, you're gonna need a persistent connection to the huge databases that comprise the AI's data. I guess it might be possible to include a trimmed DB in each standalone game, but it will be a worse experience, bigger install sizes, and a pain for devs.

 

[Mercador]

They already using it for customer support so for NPCs in a game seems a given in my book. You can limit the model to the data you want but it might be tricky to limit the answers you got.

 

[Fafalada]

If it ever does, I think that will likely mean permanent "always online" single player games.

Only if you can convince people to pay permanent subscription for single-player games.

 

[MrRibeye]

This was the video that convinced me of the feasibility and entertainment value of AI for NPCs.

 

[El Muerto]

Mantella - Bring NPCs to Life with AI

Bring every NPC to life with AI. Mantella allows you to have natural conversations with NPCs using your voice by leveraging Whisper for speech-to-text, LLMs (ChatGPT, Llama, etc) for text generation,

www.nexusmods.com

 

[T4keD0wN]

I guess it might be possible to include a trimmed DB in each standalone game, but it will be a worse experience, bigger install sizes, and a pain for devs.

The platform holders could probably begin to offer the tools, it would lessen many downsides if theyre integrated within the OS and shared across multiple games.
Wouldnt be surprised if we are one or two gens away from MS, Sony and maybe even engine and hardware makers from providing the tools for implementing AI chat into games.

 

[Pejo]

The platform holders could probably begin to offer the tools, it would lessen many downsides if theyre integrated within the OS and shared across multiple games.
Wouldnt be surprised if we are one or two gens away from MS, Sony and maybe even engine and hardware makers from providing the tools for implementing AI chat into games.

Yea that's a good point. Including a separate processor just for AI (like MS's Win11 Copilot+ machines) and offering console level integration/API would take the sting off quite a bit for devs.

But you can be damn sure that your PS+ or whatever cost is going to go up.

 

[Dorfdad]

All for it if the ai understands the logic or the game and what’s happening so far it could offer some real immersion into the world. Currently most quests are just silly conversations with no depth. Ai could really improve most games dialog

 

[{insert code here}]

It's going to be awhile. LLM's are GPU intensive and devs seem more interested in using every watt of GPU power strictly on graphics. When it happens it'll probably be some 2D indie-game that gets it right. Maybe it's an LLM-backed party member you have, or the people in a central-hub/town backed by the LLM. It certainly won't be a AAA game that does it first, if ever. They could offload all the token-generations to the cloud but then they would need a massive GPU farm to do that.

 

[Pegasus Actual]

Sooner the better.

 

[Magic Carpet]

Soon kids will never need to learn to read.
I can finally be rid of games with gibberish speech and text.

 

[elhav]

If it ever does, I think that will likely mean permanent "always online" single player games. If you want the best quality AI, you're gonna need a persistent connection to the huge databases that comprise the AI's data. I guess it might be possible to include a trimmed DB in each standalone game, but it will be a worse experience, bigger install sizes, and a pain for devs.

This is a very interesting point. I won't pretend I know anything about files compression, and I guess it's much more difficult to compress sound files, but maybe that's a technology they'll have to improve to have that huge database on a disc. Or at least a decent part that allows some generative dialog here and there.

Because I don't think requiring "always online" would go well with most people unless the game/franchise is really popular

 

[TheAssist]

not really for the main story, but you could write personas for smaller side quest npc's or merchants and stuff. Lets say you write a merchant persona and prompt it to generate 20 to 30 words every time you want to buy something. You can reuse these basic personas for lots of NPC's but you never need to write the actual dialog. But you need to feed the Ai with a lot of background information about the world and how people talk. Side quest NPC need more complex personas. Sure, the Personas could be written by AI as well, but at some point a human still has to write something so the AI can play of of it. Another cool feature would be the random "street chatter" that you hear. I am currently playing like a dragon infinite wealth and you hear the same dialog like 5 times in a row when you walk by someone. This is were AI generation makes sense. Not for main story cutscenes though.

Also, if you want the dialogue to be highly adaptive (eg reacting to things you have already done, and stuff you are wearing or NPC's in your party) the prompts become more and more complex and some point the prompts themselves become so complex, that you might as well just write the dialogue. Which might then be spoken via an Ai voice.

 

[Pejo]

This is a very interesting point. I won't pretend I know anything about files compression, and I guess it's much more difficult to compress sound files, but maybe that's a technology they'll have to improve to have that huge database on a disc. Or at least a decent part that allows some generative dialog here and there.

Because I don't think requiring "always online" would go well with most people unless the game/franchise is really popular

I think generationally, kids in the future are not going to put up the fuss that the (generally) older population of GAF does, about "always online". To them, everything is already online, so why not games too? I think it's mainly gamers age 30+ that push back against the online requirements. Especially if the requirement does something cool like actual NPC conversations and stuff, I think people will be fine with it.

Not saying it's a good thing or not, but that's just how it is.

 

[Guilty_AI]

There's more to dialogue than just words forming cohesive sentences you know. AI dialogue, even if convincing, is mostly empty and serves no purpose other than background noise. Its the kind of stuff that sounds cool on paper but doesn't really improve the game in ways that matter.

 

[elhav]

I think generationally, kids in the future are not going to put up the fuss that the (generally) older population of GAF does, about "always online". To them, everything is already online, so why not games too? I think it's mainly gamers age 30+ that push back against the online requirements. Especially if the requirement does something cool like actual NPC conversations and stuff, I think people will be fine with it.

Not saying it's a good thing or not, but that's just how it is.

Sure, in the distant future it does make sense. In the foreseeable future I think the older gamers are a big enough chunk of the customer base to make such a practice problematic

 

[ReBurn]

10 years for proper AI sound. Currently AI is dogshit

Current models are more than sufficient to create NPC tracks, both to write them and voice them.

 

[Fafalada]

I think generationally, kids in the future are not going to put up the fuss that the (generally) older population of GAF does, about "always online". To them, everything is already online, so why not games too? I think it's mainly gamers age 30+ that push back against the online requirements.

Gamers opinions are not the problem here.
The only reason online is as limited as it is today has always been down to costs - if there was a commercially viable way to either get those cloud resources for free - or have people cough-up 5$ per game per month - every game would have been online only yesterday.

 

[violence]

Games should talk less.

 

[feynoob]

Current models are more than sufficient to create NPC tracks, both to write them and voice them.

If you want robotic sound, maybe.

 

[Men_in_Boxes]

There's more to dialogue than just words forming cohesive sentences you know. AI dialogue, even if convincing, is mostly empty and serves no purpose other than background noise. Its the kind of stuff that sounds cool on paper but doesn't really improve the game in ways that matter.

This was my position a few days ago. Then I played around with ChatGPT and realized we're waaaay past that.

 

[Singular7]

I'm convinced that 90% of everything on the internet is already generated via "AI"

 

[Guilty_AI]

This was my position a few days ago. Then I played around with ChatGPT and realized we're waaaay past that.

like i said, there's a huge gap between convincing dialogue and engaging dialogue. If you're walking around virtual New York city and find two npcs having an AI generated, highly convincing conversation about the weather, you'll say "wow, cool" and promptly move on.

 

[feynoob]

This was my position a few days ago. Then I played around with ChatGPT and realized we're waaaay past that.

A robot is still a robot. Voice isnt just saying some words. It has a melody, something which your ear can pick it up.
However, robots dont have that. They cant form a melody, as they are simply reading a word.

 

[Men_in_Boxes]

A robot is still a robot. Voice isnt just saying some words. It has a melody, something which your ear can pick it up.
However, robots dont have that. They cant form a melody, as they are simply reading a word.

You and Guilty_AI need to mess around with those programs a bit. We're well past "just saying a few words". The AI can essentially have a free form conversation with you at this point. This wasn't even a topic of conversation a few short years ago. The amount of progress that will take place over the next 12 - 24 months will be crazy.

 

[feynoob]

You and Guilty_AI need to mess around with those programs a bit. We're well past "just saying a few words". The AI can essentially have a free form conversation with you at this point. This wasn't even a topic of conversation a few short years ago. The amount of progress that will take place over the next 12 - 24 months will be crazy.

AI CANT REPLICATE REAL VOICE.

Voice is not something a machine can create nor understand how the words are being said. It simply samples the voice, then tries to do a lip service. But its still a robot voice.





You can still detect something is wrong with the voice. There is no pause or any movement. You can do tricks by shorting the words and adding music to make it real though, but still doesnt work without those.

 

[Comandr]

AI CANT REPLICATE REAL VOICE.

Voice is not something a machine can create nor understand how the words are being said. It simply samples the voice, then tries to do a lip service. But its still a robot voice.





You can still detect something is wrong with the voice. There is no pause or any movement. You can do tricks by shorting the words and adding music to make it real though, but still doesnt work without those.

Your first video you linked seems to defeat your own argument. Deepfake Cooper says at the end “90% of people cannot tell the difference.” I use AI tools pretty extensively and if I didn’t know I was listening to a fake version of him, I wouldn’t have thought about it for a second. At some point, it’s good enough.

 

[Guilty_AI]

You and Guilty_AI need to mess around with those programs a bit. We're well past "just saying a few words". The AI can essentially have a free form conversation with you at this point. This wasn't even a topic of conversation a few short years ago. The amount of progress that will take place over the next 12 - 24 months will be crazy.

Like i said, just having a convincing conversation, no matter how long winded or how realistic the voice sounds, isn't enough. There's far too much nuance in story-telling, especially in games, these conversation AIs can't pick up or generate on their own. Things such as nuanced emotional state, motivations, personality, way of expressing oneself, foreshadowing, pacing, story-structure, author's personal flavor and so on.

 

[farmerboy]

Being able to have a convo with an npc ingame and have it understand you is a fascinating concept I hadn't thought about.

 

[Comandr]

Like i said, just having a convincing conversation, no matter how long winded or how realistic the voice sounds, isn't enough. There's far too much nuance in story-telling, especially in games, these conversation AIs can't pick up or generate on their own. Things nuanced emotional state, motivations, personality, way of expressing oneself, foreshadowing, pacing, story-structure and so on.

How about one voice actor clearly doing multiple roles, or every guard in Skyrim mentioning they used to be an adventurer like you in the exact same tone and inflection - because it’s the exact same voice line.

There are plenty of places that AI voice and generative dialog could be applied convincingly and appropriately that would ONLY enhance the situation.

 

[feynoob]

Your first video you linked seems to defeat your own argument. Deepfake Cooper says at the end “90% of people cannot tell the difference.” I use AI tools pretty extensively and if I didn’t know I was listening to a fake version of him, I wouldn’t have thought about it for a second. At some point, it’s good enough.

90% of people simply ignore it. They can tell the difference, if they pay attention to the sound.
In an enviroment which you are paying attention to the voice, it doesnt work. Your attention picks up the eratic voice of the AI.

 

[Comandr]

90% of people simply ignore it. They can tell the difference, if they pay attention to the sound.
In an enviroment which you are paying attention to the voice, it doesnt work. Your attention picks up the eratic voice of the AI.

You’re just twisting the narrative to fit your argument. I literally admitted that I wouldn’t have questioned it if I wasn’t actually looking at the screen. That wasn’t me willfully ignoring anything. I couldn’t tell. Most people when they hear throwaway dialogue from a given npc don’t sit and pour over the vocal details.