General Chat GPT AI BOTS\NPC

[Marcus Bronco]

As a Mod that for unstable Mount and Blade by Tales Worlds (very contested) does, this game ED would be perfect, especially for Odyssey but also not odyssey players should access it, a Chat GPT system that allows you to talk with NPC and receive "illusion" they know stuff (of course each NPC should have a scheme of coding that programmers set in order for that NPC "not to know some stuff" as maybe they are from far away or it's impossible they answer you like they know 100% of data, they are not gonna feel real if they answer like a Search engine!
But this would be the only work..
Rest is very "simple" to program, they just feed the entiure machine by SECTORS of knowledge \ chapters stuff (I say chapters because of the thing just explaiend above on not knowing everything each NPC like this respecting the CONTEXT of who they are where they are how much they could be knowing ...) and then feed this to the various NPC in a procedural way.. and you will be able to write stuff and talk to them..
Maybe even via audio, who knows with speech recognition.. but this would be a lot a lot.. as the game already does stuff in this line (Great design of Audio like even the Radio modifier effect and various static \ distortions, so I would "not be" surprised if also this layer of Chat GPT was inserted: one that allows to talk to it and get answers .. like we do with HCS packages and voice attack.

Anyway... Inworld system or just a propietary pool of data, as it should be anyway happening would really be cool as long as it doesn't take away too much time or effort from other more important things to develop.

I think this game could really be one of the longest surviving ones. I think they should invest on it.

---

 

[Screemonster]

God, NFTs and throwing GPT in where it isn't needed? It's like a techbro buzzword soup in here.
Survival mechanics in an always-online live-service game? You want to add a SLEEP counter? IN A GAME THAT ISN'T PAUSED WHILE WE, THE PLAYERS, ARE ASLEEP?

God, if I didn't know better I'd think you threw the open-only BGS thing in there with all the rest of this mess specifically to discredit it.

 

[Nill]

About the NFTs and removing BGS: what kind of game assassination is that?
About language model AI: For what purpose? Will it just generate text without leading to any action/consequence/mission?

 

[Fizzatron]

AI Role Playing NPC interaction is likely to become standard at some point. Not all games would benefit from it but Elite would, IMO. Being able to waffle to your crew or a rando in the concourse etc would be pretty neat, especially if they were heavily trained on Elite lore, you could interview your potential crew even, as the AI could generate NPCs with mood traits and things like that.

I also like the idea of eating/drinking but I think it would only work as a cosmetic effect, in a social way, kind of how drinking is in Deep Rock Galactic, say. With crafting it would be a good way to incorporate some resource gathering for recipes etc.. Empyrion has this and it's pretty cool to have a stocked fridge.

Sleeping isn't necessary, we can all imagine our avatars falling asleep when we log off, though I would like to see some Captain's Quarters on fleet carriers that had a bed we could retire to if so wished.

I don't think NFTs need to be a thing, but going back to the AI bit there could be an NPC outfitter who might be able to create unique decals based on prompts and where in the occupied space they reside, which could be cool. Being able to share them to your squadron would be good too.

 

[Rainbro]

For the effort and cost of making it work to a reasonable standard, you could pay a handful of people to generate thousands of fitting NPC statements. But hey, "pay a corporation for an over-engineered solution" is always more attractive to execs than "just do the actual work".

 

[Felix DiCestria]

As a Mod that for unstable Mount and Blade by Tales Worlds (very contested) does, this game ED would be perfect, especially for Odyssey but also not odyssey players should access it, a Chat GPT system that allows you to talk with NPC and receive "illusion" they know stuff (of course each NPC should have a scheme of coding that programmers set in order for that NPC "not to know some stuff" as maybe they are from far away or it's impossible they answer you like they know 100% of data, they are not gonna feel real if they answer like a Search engine!
But this would be the only work..
Rest is very "simple" to program, they just feed the entiure machine by SECTORS of knowledge \ chapters stuff (I say chapters because of the thing just explaiend above on not knowing everything each NPC like this respecting the CONTEXT of who they are where they are how much they could be knowing ...) and then feed this to the various NPC in a procedural way.. and you will be able to write stuff and talk to them..
Maybe even via audio, who knows with speech recognition.. but this would be a lot a lot.. as the game already does stuff in this line (Great design of Audio like even the Radio modifier effect and various static \ distortions, so I would "not be" surprised if also this layer of Chat GPT was inserted: one that allows to talk to it and get answers .. like we do with HCS packages and voice attack.

Anyway... Inworld system or just a propietary pool of data, as it should be anyway happening would really be cool as long as it doesn't take away too much time or effort from other more important things to develop.

I think this game could really be one of the longest surviving ones. I think they should invest on it.

---

Did an AI write this post?
Try searching youtube for AI generated Country songs for examples of why the answer is NO!!!

 

[Screemonster]

For the effort and cost of making it work to a reasonable standard, you could pay a handful of people to generate thousands of fitting NPC statements. But hey, "pay a corporation for an over-engineered solution" is always more attractive to execs than "just do the actual work".

Personally I don't think anyone should be trying to bring up computer-aided asset generation to David Braben of all people or suggesting it for Elite as a franchise in particular

Dude practically invented procgen in gaming, if anything it's a little insulting to "suggest" it as if the devs hadn't thought about the possibility of having... an algorithm... design... the game.
And part of the craft is knowing exactly what parts should be generated and what parts should be written.

 

[Fixitman]

"Modern" AI is nothing but a marketing gimmick, and it has no place in any serious game.

 

[Fizzatron]

For the effort and cost of making it work to a reasonable standard, you could pay a handful of people to generate thousands of fitting NPC statements. But hey, "pay a corporation for an over-engineered solution" is always more attractive to execs than "just do the actual work".

I feel like we're at a bit of a psuedo 3d vs real 3d point with NPC AI. A lot of the advantages of creating a library of sounds vs actual AI work in the short term but with the AI, it could respond directly to situational contexts in ways that trying to patch together dialogue to fake that wouldn't achieve. But I get your point, the resource costs are still a major factor but I think this will come down more over time.

I would also imagine that once an AI is trained on a range of base character traits, being able to mix them up to create more complex combinations of traits that are more representative of a real person would open up the possibilities of generating an infinite resource of NPCs that are 'aware' of their traits and surroundings. Imagine having accents from specific regions of the bubble that could be identified once you become accustomed to them. There's lots of possibilities.

 

[MUTILADOR]

As a Mod that for unstable Mount and Blade by Tales Worlds (very contested) does, this game ED would be perfect, especially for Odyssey but also not odyssey players should access it, a Chat GPT system that allows you to talk with NPC and receive "illusion" they know stuff (of course each NPC should have a scheme of coding that programmers set in order for that NPC "not to know some stuff" as maybe they are from far away or it's impossible they answer you like they know 100% of data, they are not gonna feel real if they answer like a Search engine!
But this would be the only work..
Rest is very "simple" to program, they just feed the entiure machine by SECTORS of knowledge \ chapters stuff (I say chapters because of the thing just explaiend above on not knowing everything each NPC like this respecting the CONTEXT of who they are where they are how much they could be knowing ...) and then feed this to the various NPC in a procedural way.. and you will be able to write stuff and talk to them..
Maybe even via audio, who knows with speech recognition.. but this would be a lot a lot.. as the game already does stuff in this line (Great design of Audio like even the Radio modifier effect and various static \ distortions, so I would "not be" surprised if also this layer of Chat GPT was inserted: one that allows to talk to it and get answers .. like we do with HCS packages and voice attack.

Anyway... Inworld system or just a propietary pool of data, as it should be anyway happening would really be cool as long as it doesn't take away too much time or effort from other more important things to develop.

I think this game could really be one of the longest surviving ones. I think they should invest on it.

---

Without a doubt, what you mention will be the trend from now on. If the elite join this trend, it would be a smart move, if they don't, others smarter will do it. At the moment you can use GPT while playing Elite, I use it to talk about possible missions while piloting my ship or something else.

Source: https://www.youtube.com/watch?v=vWMLVZF3pGc&t=118s

 

[JCCarven]

"Modern" AI is nothing but a marketing gimmick, and it has no place in any serious game.

Don't know, pal.. You dealing with absolutes like a Sith. Seems like it will be a common thing in the future. Maybe not so near future, but I could easily imagine lightweight and simple language model working on device like.. PlayStation 8 or something?.. You just wait until PC will go ARM as Apple did.

 

[Nill]

Without a doubt, what you mention will be the trend from now on. If the elite join this trend, it would be a smart move, if they don't, others smarter will do it. At the moment you can use GPT while playing Elite, I use it to talk about possible missions while piloting my ship or something else.

Source: https://www.youtube.com/watch?v=vWMLVZF3pGc&t=118s

This is just.... extremely strange to see these kind of interactions, AI doesnt know how act in characters ~human like~, it seems that everyone is a robot constantly saying the same things with the same personality. Indeed it would be cool otherwise. But not as it is right now, just doesnt fit in anything well.
And no, AI isnt the future of "NPC Interactions", games and NPCs are much more than that, okay, we maybe will see a new classification of games of GPT-NPC, but not my thing right now.

Also, the kind of interactions the player do in ED isnt a normal-like interaction of an story game. Thats not how ED players interact with the lore. Its via BGS.

 

[Ian Doncaster]

Another consideration is cost - Microsoft charges about a cent per query for Copilot in the bulk-pricing modes, and I expect other chatbot services are similarly priced if you're not on a free "personal use" tier (that would have nowhere near the query limits to run a game of this scale on, of course). And this is at the point where the chatbot creators are all offering them at a substantial loss/subsidy to try to get people hooked and willing to pay commercially-viable rates later.

So in a typical hour of play I might interact with 20-30 mission offers (with more messages needed for the ones I take and complete, especially if they come with wrinkles), plus see background chatter from at least as many passing ships. If all of those are being generated by directly querying a chatbot service then that's cost Frontier most of a dollar.

I don't pay Frontier a dollar per hour to play the game, and even if I did Frontier won't want to spend that dollar on generating text that I'm barely reading in the first place. Even if they keep mission text and space NPC chatter to static templates, so you have to do something separate to interact with the chatbot, the costs could quite rapidly add up - compared with what people pay for the game in the first place! - if anyone actually used it.

That's the thing that's going to make "AI NPCs" completely impractical for anything other than a "mod the game and plug your own free/heavily-subsidised ChatGPT account into it" approach, at least until someone figures out a way to go from what we currently have:
"the English language is sufficiently statistically regular that grammatically correct text on a specific theme can be generated programatically"
to:
"and we've figured out a smarter way to do it than throwing a ridiculous amount of computing power at it"

 

[JCCarven]

That's the thing that's going to make "AI NPCs" completely impractical for anything other than a "mod the game and plug your own free/heavily-subsidised ChatGPT account into it" approach

Obviously, in the current state, this is a dumb idea. Brave, but dumb. Language models should be running on-device. Small, fast, and trained for a specific purpose. I don't see why someone should seriously discuss using commercial LLMs in the cloud for gaming. They are too general for this stuff.

 

[Ian Doncaster]

Language models should be running on-device. Small, fast, and trained for a specific purpose.

And at that point you can certainly do the same thing a lot quicker and cheaper - in terms of what's actually needed to make the game work - by just writing the text the game needs the old-fashioned way. Sure, you won't be able to have conversations on an arbitrary topic with an NPC, but for Elite Dangerous that's a very good thing because it would reveal just how much of the game world is entirely undefined or abstracted to start with [1].

(Especially since for Elite Dangerous you'd need to train and verify five language models, for each of the supported languages - and that means bringing in programmers and testers who know those languages well, rather than just sending some flat strings to a translation agency.)


[1] For example, "who is the Alliance Assembly representative for Ross 1015?" is a question with a definite and public in-universe answer that, at the very least, any moderately informed NPC in the Ross 1015 system should be able to tell you. It's also information that Frontier likely hasn't defined - and I picked the superpower with one rep per system deliberately, the Federal and Imperial structures are more complex - because it's completely irrelevant to anything we do as CMDRs. So the options are:
- spend a huge amount of time defining all this level of background information [2] so that it's available to be used in answers
- let the chatbot make up a different answer every time someone asks, or implausibly claim not to know who the representative is while being able to tell you instantly what Der's argument of periapsis is
- (easiest) don't let the player ask that in the first place, which means not allowing arbitrary questions, which means a conventional hand-written dialogue tree is absolutely fine

[2] Sure, running their NPC name generator 1200 times and saving the results so that this specific question has a defined answer isn't a big deal. But there's thousands of similar things out there.

 

[Rainbro]

Obviously, in the current state, this is a dumb idea. Brave, but dumb. Language models should be running on-device. Small, fast, and trained for a specific purpose. I don't see why someone should seriously discuss using commercial LLMs in the cloud for gaming. They are too general for this stuff.

I don't think this is feasible. If there is a small, fast model that can generate good responses, why wouldn't the cloud-based companies be using it? And them being "general" isn't something that can be solved - the bulk of the work of understanding language is the same regardless of the context, unless we want to make a model that "just" knows Elite and spits out sentences like "Asp Explorer Federation Coriolis". That context can only be in addition to the "general" models.

 

[JCCarven]

Sure, you won't be able to have conversations on an arbitrary topic with an NPC, but for Elite Dangerous that's a very good thing because it would reveal just how much of the game world is entirely undefined or abstracted to start with

Sorry for the misunderstanding. I didn't mean Elite specifically, I was speaking in general.

If there is a small, fast model that can generate good responses, why wouldn't the cloud-based companies be using it?

But there are, and they are even open source. But I guess you haven't seen the papers on mini-Orca. They are certainly not as good as GPT-4 and won't be able to write you a code for a Flappy Bird analog, but they can handle basic conversations, and you can train them with just synthetic data. Siri's LLM will be run on-device for basic things, or at least this is what Apple told us.

Just give it some time (several years) and some Game-LLM Company logo will be on every loading screen like Wwise, BINK, etc.

Once again, I think clarification is needed: I'm not talking about Elite specifically, especially in current state.

 

[Rainbro]

But there are, and they are even open source. But I guess you haven't seen the papers on mini-Orca.

I don't consider 4-8GB of RAM, plus GPU resources, for the weakest model, to be "small" - and thats giving it the benefit of the doubt that it would even run well under those constraints. If the "background NPC chatter" module is taking up resources on the order of magnitude of the entire rest of the game, something has probably gone very wrong. Thats like several Odyssey's worth of system requirement increases, and we know how well that went down.

 

[JCCarven]

I don't consider 4-8GB of RAM, plus GPU resources, for the weakest model, to be "small" - and thats giving it the benefit of the doubt that it would even run well under those constraints. If the "background NPC chatter" module is taking up resources on the order of magnitude of the entire rest of the game, something has probably gone very wrong. Thats like several Odyssey's worth of system requirement increases, and we know how well that went down.

Cmon, dude, have you read this?

Once again, I think clarification is needed: I'm not talking about Elite specifically, especially in current state.

I'm speaking about technology in general (IN GENERAL — I said it several times) that definitely will be there in gaming in several years. Cause some people saying that this is scam and vaporware. And you keep posting about chat-gpt overrunning your background NPC chatter..

 

[Rainbro]

Cmon, dude, have you read this?

I'm speaking about technology in general (IN GENERAL — I said it several times) that definitely will be there in gaming in several years. Cause some people saying that this is scam and vaporware. And you keep posting about chat-gpt overrunning your background NPC chatter..

You gave Orca as a current example of a "small" LLM, and it uses resources comparable to the entire game. It would be nice if some future LLM magically solves the issues assosciated with them, but that's baseless speculation.

In the meanwhile, Frontier could just write some more dialogue rather than waiting an undefined number of years for an insane hypothetical breakthrough that makes this approach viable. I am not convinced by the idea that this is inevitable - every salesperson says their product will inevitably be the next big thing.