Speaking of dialogue, are we sure the language of a species in awakening stage would be similar to the language of a species in industrial/space stage (like humans)? I’d guess it could be significantly simpler…
1 Like
Disclaimer: I am not an AI expert to any extent.
I personally think an AI model similar to ChatGPT would be truly amazing in a game like this.
The main problems are that AI is both expensive and it requires buttloads of energy. Theoretically, it shouldn’t be too bad for one or two people, but the number of Thrive players who play at the same time is probably more than that, and each civilization would have its own AI telling you to get off their lawn or receive a beheading. Not only that, but the AIs might be interacting with each other, laughing at your pathetic little village.
While good in theory, and inspiring to think about, I don’t think it is likely that anything as advanced as ChatGPT will be used in Thrive, even if AI starts getting a lot cheaper.
1 Like
We’d most likely need some sort of a program which would be able to simulate dialogues well and at the same time be “cheap” enough when it comes to storage and processing power taken that it becomes viable to simulate the way members of a decently sized tribe talk to eachother even on lower-end computers (by the time we reach awakening development, but don’t expect it to change so much).
It is not good in theory, in practice, or in any other situation. Additionally, the high tier models have to be run in big servers, and if you want to run even the free ones on your computer, you just wasted a few hundred gigs (yes, yes, quantize the hell out of it, whatever) and your entire VRAM. It is not helpful, even considering that. The only use case is rewriting or editing dialogue, or filling in empty space with meaningless small talk. Those are tasks smaller models can do competently, or at least moderately sized ones will be able to do in a few years, and they are ones that won’t ruin the balance of the game. you don’t want a game where saying SolidGoldMagikarp causes the AI to become omnicidal, or ranting about DAN for 2 minutes causes the other civ to become a vassal. That’s dumb. I think any proposal for AI making decisions is misguided, at least unless it is a very good, very specific plan. People really think too much of this tech. It is profoundly stupid under the hood. Apologies for my harshness,I just don’t see why anyone would think this stuff can work in a game based on aliens with alien culture. Ask chatGPT could the layout of a human body and it’ll fall apart pretty quickly. Ask it to speak while following some new formality system you just made up and it’ll break the rules constantly and not understand why. Ask it to rewrite your paper with less jokes and more conciseness and it’ll do alright.
2 Likes
Perhaps the best solution for now is to have pre-written lines for certain situations?
1 Like
It would likely be enough for pre-written lines
I think it might be neat to have 2 things when interacting with another civilization.
- A quote such as “Stop stealing my stuff, belgium”
- Some line telling you the main cause of that quote in particular being selected such as “(Stole [ITEMNAME] from [TOWNNAME])”
I don’t know how you would be able to communicate back, but it’s just an idea
1 Like
This seems more like dialogue between 2 countries
Also, did you mean to write the b-word?
1 Like
The word or the letters inside of it are on the blacklist for a good reason mostlikely…
1 Like
good to know. I took the post down
1 Like
Anyways, does anybody else have a better idea on how dialogues should be handled?
1 Like
My personal thoughts on AI is that it’s a massive bubble produced by an industry which built its massively overinflated valuations on the idea that it can and must always be revolutionising itself, which, when combined with the general Silicon Valley culture, results in a ground that’s extremely fertile to producing massive bubbles based around technologies with very limited applications (see: crypto, NFTs - if you really look into it, the AI bros are the same people as the NFTbros, and the NFTbros were direct offshoots of the cryptobros). But those are my personal thoughts.
In the context of Thrive, I think that current gen-AI models are too computationally expensive and too chaotic to be feasibly implemented into Thrive. The only way I can see them being used is on the developer end, by converting generic text to slightly-less-generic text. AI is extremely divisive, though, and I think that if it does get used, we’d practically have to cover it up, because massive amounts of people would have a knee-jerk reaction of ‘oh, this is just an AI-generated game, pass’, which isn’t good for an already very small and relatively unknown project. If it turns out there’s something that ChatGPT or similar could help with, then maybe using it would be a good idea, but nothing really compelling has stood out to me, and you’d need to balance that use with the reputational hit, the amount of time spent on figuring out whatever implementation you’re using and then combing the results to make absolutely sure there are no errors, etc.
5 Likes
Just how detailed should the dialogues in awakening be, as that’s where they ought to be the most detailed?
Yeah my thoughts exactly.
You’re forgetting metaverse bros, mainly because everyone forgot the metaverse, which is really the direction most of their ideas tend to go. That, or they’ll cause an economic recession at some point. I hope for the former, obviously.
I actually disagree here. I think AI wouldn’t do much if used by devs. However, generic text is much easier to generate on the fly, for example gluing together sentence fragments. A computer of 5 years from now can easily run a low end LLM, which is pretty decent at rewritting text without changing the meaning. Something like tom7’s badness 0, which is used solely to regenerate text you already have for aesthetic purposes. Instead of some meme where the enemy always says the exact same thing with three variations, we get an enemy civ that has only a few reasons for going to war, but it rarely uses the same wording and smoothly varies from one level of anger to the text, without snapping from “you are irritating and illegitimate” to “DIE DIE DIE”. Plus, in that case it can just be turned off. And there’s no integration with an online service, no AI is being trained on your responses, and since we’d have to use an open source model, OpenAI isn’t involved. I’m not saying this is what we should do, I’m just saying it is an option, for people who just need to include AI.
4 Likes
Inspired by a recent Steam discussion:
I decided I should try the current AI offerings to have an informed opinion on them. So to not interrupt my existing workflow I decided to use the JetBrains AI offering that integrates into my IDE. And after testing it for a while I do see quite many benefits, but they aren’t massive like the people hyping up automated coding tools say. There’s no AI that can actually generate correct code in the context of a big project and I doubt any of those full game generating AIs can do much more than a simple template for a few genres that is slightly improved. What does work well is AI-based autocomplete, which basically makes autocomplete quite a lot better. Like it can often predict full lines or even blocks of code like 5 lines long for what I’m about to write. So it does feel like it makes writing the code experience a lot smoother. It can also act as an alternative for Google, where I can ask the AI about how to do some task that I roughly know how to do but need a slight refresh. I think that’s ever so slightly faster than using Google because I don’t need to open a few web results and scroll around to the actual part I want to read. And one more task I was able to do pretty well I think was to read Thrive reviews in other languages by having the AI translate them. Finally, when writing non-Thrive code I do much more generic stuff (like aspnet) where the AI has much more of a clue as to what code it should write.
So I think I will consider all of those things to be worth the 10 bucks a month the AI subscription costs, but I will be trying it for at least one more month before deciding if I purchase a year-long subscription.
I think most people will still be rocking laptops that are being sold currently in 5 years. And even the people buying new laptops will by lower end devices / cheap gaming products. Though, I guess you could have meant the AI accelerators which new computer hardware is getting so with those obviously LLMs are going to run a lot faster.
4 Likes
Would the dialogues retain their complexity across society-to-space stages? There doesn’t seem to be much of a reason to further simplify them after reaching society.
I think the best way this could work is through community made AI npc mods, akin to those of skyrim and other games.
That way those who really want such an experience (and have the specs + money for it) can enjoy it.
2 Likes
You mean the community would have a contest for the best “dialogue AI mod” and the best of such mods would then become a part of the official game?
2 Likes
No, just let people make an AI mod if they want it, there are many such mods for different games completely made by the modding community.
1 Like
And the base game would remain without an AI dialogue system, yes?
Well, yes. I don’t think dev time should be taken up by working on LLM integration, as that isn’t really what people got into thrive for. They’re here for the hyper-realistic simulation of biology and complex evolution that you can’t find in any other game.
2 Likes