Realistic dialogue

I really don’t think that adding ChatGPT to Thrive is a good idea. It’s unnecessary and too expensive. AIs can’t even parse backwards text, much less think, or be useful gameplay-wise. Don’t think this should be implemented. It would be easier to just have the devs write out a bunch of dialogue. Depending on how easy it is to implement dialogue, I could write dialogue (that’s dependent on the species personality). No need for AI.
Also it’s a buzzword

3 Likes

Honestly, the biggest problem with ai is, as hhyyrylainen said, would be trying to translate what the ai says to the rest of the game. You’d have so many glitches where the ai says one thing and the game does another, or it gets out of character, or if we’re talking about chatGPT here, it might be way too easy to gaslight. It doesn’t really matter how realistic your dialogue is if the actions taken because of it are not.

Honestly, I don’t like ai very much. It’s shoved into my face way too often. I’d prefer handmade options, but it’s whatever the devs decide. So whenever we get to that point as long as it’s done well, I’m good. Though it definitely should be a part of dlc or something instead of the main game.

3 Likes

Ai for games is an awful idea imo, unless the whole game is built around it. Llms do not understand anything outside the context of the conversation and you can do a prompt injection to convince an AI to order it’s people to eat their favorite children. That’s a dumb idea.

I don’t know if anyone mentioned this but a species might talk a certain way compared to humans or might have certain expressions or mannerisms that humans don’t have which an AI as of now wouldn’t be able to most likely replicate well.

Same thing would go for individuals but would be more pronounced.

1 Like

I really hate the idea of implementing any sort of LLM into Thrive, and given how bad they are at acting like humans, I’d imagine they’d be far worse if they were trying to simulate an alien. Seriously, go to ChatGPT (or Bing) and ask it to act like an alien.

More to the point, this completely goes against Thrive’s development philosophy. It would never fly, and that’s assuming it’s even remotely feasible to implement. The only games I know of which implement AI are shovelware made by the same people who were all gung-ho about crypto games and NFT games, and they’re mostly using AI image generation to cut costs on something that was already programmed in a week.

5 Likes

fallacy of composition. Uhh, you’re otherwise 100% right. LLMs are an atrocious idea. Possible AI derived text could help turn something like a logical string of transfers of information with a local goal and a terminal goal into sensible dialog, but even that wouldn’t be best done with an LLM. AI is designed to passably imitate a human, the opposite of what you want, when deeply imitating an alien, especially with deciding things.

3 Likes

There’s now pretty good open LLM models that we could add into Thrive (clarification: what I mean with the “open” is that you can download the model and run it on your computer without having to pay any AI API fees), as long as we are fine with Thrive being only playable on desktop GPUs with 8+ GB of memory, which is a pretty huge ask (as we have many laptop players). And the LLM would still take as long to reply as the online LLMs so realtime discussions would feel slow.

4 Likes

(myself being one of those laptop users)

2 Likes

It’s also important to mention that awakening stage will only enter development in 2030s, so by that time we could have access to better tools for creating artificial dialogue.

Genuinely, I FULLY disagree. LLMs are not going to be simple to put in a laptop. People will still use laptops made today, and a modern laptop doesn’t have the ram and/or GPU to run anything bigger than a quantized 17b model or something like that. Genuinely, that is not going to get you a better experience than GPT 4o, I don’t care if we spend a decade getting better tech. if 4o is not good enough, whatever models will fit on a decent laptop won’t wither. If you are referring to some other tech, what? I can’t think of anything. Maybe we could use an LLM to rewrite very formulaic dialog that just plugs together sentences like “I want this thing from these people because” with attitude and flavor. In a few years we might get LLMs that’re really good at preserving the original meaning of a text. BUT, we can do that now!! I’ve done some locallama nonsense, you can go build a thing that does that today, and to get good results, you need some pretty high specs. What we can do on a decent PC is about what we can do in a few years on any old laptop. I cannot, for the life of me, think of any truly new ideas that could come about. If you’re just saying “technology changes, you can’t predict what will happen!” You might as well post that at the end of every single awakening, society, industrial, or space stage thread. it doesn’t mean anything and it doesn’t help. If we know what we’re waiting for we know not to stress that we have nothing, because we have something, it’s just not ready, but if we have nothing to wait for, telling people to wait is pointless.

4 Likes

I agree with you.


Would the dialogue show up only if you communicate using sounds or too if you communicate via other means?

I think since humans communicate with text and sounds we should use text for all communication. no reason to discriminate against ants or something. We could have sound effects for people who communicate with sound, but it isn’t a requirement like subtitles are imo.

3 Likes

How would dialogue work with eusocial colonies? Would it still be present as communication between different hives?

1 Like

I really need to do more research into naked mole rats because they aren’t bugs, and ants/bees and whatnot communicate mostly with lingering pheromone trails that either grow or fade as the intel they represent changes in accuracy. If you use antlike communication I don’t think subtitles make much sense as you aren’t conveying information to a target in a conversation, you’re just signposting facts, or in the case of propaganda pheromones just saying random nonsense so ants don’t know what to trust. Maybe like, dark souls bloodstains on the floor. I dunno. But yeah, I’ll have to look into naked moles rats. Plus, bee dances would totally make sense as language with subtitles and whatnot.

2 Likes

So, when you are an eusocial queen, would you have your lesser castelings talk to you and based on what they told you, you’d give them orders?

I mean. on earth the queen isn’t the boss. The queen makes eggs. The males fertilize the eggs. The drone coordinate with each other’s pheromones. No one bosses anyone around, If a be dances and says “go that way, lots of flowers” other bees can just ignore it freely, but they like helping and know flowers are good and other bees tend to be trustworthy.

2 Likes

Perhaps you’d regulate how a hive as a whole acts by the behaviour editor…

2 Likes

At first glance, this may seem like a good feature, but it creates many problems: 1. the neural network can “hallucinate” giving bad and incorrect answers. 2. When the scale of the game increases from individual communities with 5-20 members to states with many millions of citizens, the load on the computer can become a problem. 3. It requires a lot of resources that could be spent on more useful things. 4. From a game design perspective, this might work for colony sims (like rimworld and dwarf fortress), but it’s clearly not suitable for thrive.

1 Like

Responding to 2, this is why mechanics change over the game. You can’t simulate cells in aware stage like how you did in microbe stage. Same could apply here.

As I said earlier, this sounds like a terrible idea for me. There are legitimate uses for modern AI tools - I’d say science and medicine, at least - but, effectively, they find patterns. They can’t generate anything, they can’t make anything new. Summarise a database, predict the waves, digitally fold proteins - those are viable uses for LLM AI. (At least, I think stuff like AlphaFold works on LLMs. I don’t actually know.)

At best, as Deathwake said, you could use AI to make formulaic dialogue. The Infinite Fusion guy used it for the pokedex entries, I think. But it’s doubtful you could use it for anything more, and the dev time wouldn’t be worth it.

1 Like