[Suggestion] Expand your mechanics. GPT Chat + Voice.

[Rc2F6]

I saw in one video how a guy bolted a GPT-3 to his VR game, bolted on a voice generator. He would ask questions through a microphone, the voice would be translated into text using a script, the text would be sent to GPT-3, GPT-3 would get the information the character needed to know, from that information GPT-3 would generate an answer based on the question and send the text, the text would be voiced by the voice generator and the NPC would respond. If such mechanics existed, the game would be the first game to use such mechanics. The video is called [OpenAI GPT-3 Powered NPCs: A Must-Watch Glimpse Of The Future (Modbox)].

P.S. X jester gave me X awards 😆.

[Hollow]

Right now its probably not possible. In order to do what you're suggesting you would need to use an API key for chatgpt and you have to pay per token. so essentially every word it generates for everyone would cost them money

[zelosdomingo]

You could attempt to do it with one of the open source local LLMs. They do okay with non-coding stuff. Would have nutty system requirements, though. Honestly, wouldn't really be THAT hard, actually. Would certainly add a lot of variety, especially if you let the model decide what the NPC did after the conversation as well.

[Rc2F6]

Ursprünglich geschrieben von Hollow:

Right now its probably not possible. In order to do what you're suggesting you would need to use an API key for chatgpt and you have to pay per token. so essentially every word it generates for everyone would cost them money

On GitHub there is a version and GPT-3 and GPT-4, 6GB and 11 GB. These versions are not very smart, but for those tasks that are needed in the game, they can cope perfectly. For free and at least communicate with NPCs all day long.

[Masterman397]

https://www.youtube.com/watch?v=Ba7pipuRfBs

[Voelkar]

I dont think adding a GPT-3 bot would be a wise idea. With a game that focuses on precise information gathering those chatbots will definitely screw it all up by giving false information

[zelosdomingo]

Ursprünglich geschrieben von Voelkar:

I dont think adding a GPT-3 bot would be a wise idea. With a game that focuses on precise information gathering those chatbots will definitely screw it all up by giving false information

You can just spin it as misdirection, if they've got something to hide. Why people who are supposed to be neutral or helping you are saying crazy stuff would be harder to explain... People do just say weird stuff sometimes in real life, I guess.

[Voelkar]

Ursprünglich geschrieben von zelosdomingo:

Ursprünglich geschrieben von Voelkar:

I dont think adding a GPT-3 bot would be a wise idea. With a game that focuses on precise information gathering those chatbots will definitely screw it all up by giving false information

You can just spin it as misdirection, if they've got something to hide. Why people who are supposed to be neutral or helping you are saying crazy stuff would be harder to explain... People do just say weird stuff sometimes in real life, I guess.

Still, the amount of falso information given by GPT-3 vastly outweighs the correct one. I guess it wokrs kinda the same in real life but not as often as GPT-3 does with comepletely made up information that it pulled out of thin air

[Tegiminis]

Please do not put a large language model into this game.

Beyond the cost and technical issues outlined above, the fact is that LLMs simply aren't very good at generating ~actual good text~. They are okay at recontextualizing rigid information into a conversational tone, and sometimes mimicking an existing writer, but that is their limit. They are novel precisely once, then they are irritating. I find they only impress people who aren't writers, who I should point out can instantly see through the generated text to the prompt.

I have my problems with how mechanistic the social experience in this game is, but slathering over it with an LLM will not fix any issues, and will in fact just make it more annoying and less consistent to interact with the characters.

Please hire a narrative designer instead. This game has a lot of promise, but it's still lacking in the writing department. Working with a human being aligned with your goals and with their own creative energy will do more for this game than jumbling conversations up with an LLM.

[zelosdomingo]

The problem with human generated text is that it is more or less static. I would agree that shoehorning an LLM into the game would not be feasible at the moment, in any case, due to system requirements, so I suppose it's a moot discussion. Definitely would agree they need to expand upon the social elements of the game, considering that's like half of the game, if you don't like to sneak about.

[Tegiminis]

Static text is better for games for a bunch of reasons.

1) They communicate the bounds of the play space. Players associate certain phrases with certain actions (combat barks, for example) and can use that to play or make decisions predictively. Reliance on UI placement or reading comprehension is a lot more flawed than simple static text.

2) They provide a canvas for a narrative designer to paint on. Games are art, and part of the artistic process is conveying subjective human information such as views or perspectives. They are the sum total of a person's comprehension. LLMs can't comprehend, so they can't create art.

3) They help construct the "negative space" of what a game COULD be, which allows for new ideas to flourish. Humans are pattern-seeking, and as designers notice patterns in their work, they also notice anti-patterns; places where the pattern doesn't cover, or can't cover. This adds richness to the text.

Serviceably boring static text is vastly superior to infinite useless garbage text, which is what LLMs produce.

[Ted Bunny]

Ursprünglich geschrieben von Tegiminis:

Static text is better for games for a bunch of reasons.

1) They communicate the bounds of the play space. Players associate certain phrases with certain actions (combat barks, for example) and can use that to play or make decisions predictively. Reliance on UI placement or reading comprehension is a lot more flawed than simple static text.

2) They provide a canvas for a narrative designer to paint on. Games are art, and part of the artistic process is conveying subjective human information such as views or perspectives. They are the sum total of a person's comprehension. LLMs can't comprehend, so they can't create art.

3) They help construct the "negative space" of what a game COULD be, which allows for new ideas to flourish. Humans are pattern-seeking, and as designers notice patterns in their work, they also notice anti-patterns; places where the pattern doesn't cover, or can't cover. This adds richness to the text.

Serviceably boring static text is vastly superior to infinite useless garbage text, which is what LLMs produce.

1. Why is this set in stone? There's no rule that everything in a game has to be legible. Sometimes people say ambiguous things.

2. Defining what is and isn't art is a rabbit hole. There's also no rule that games have to be art.

3. I am not really sure what you're saying in this part.

As for your addendum - citation needed. I've messed around with making an RPG with ChatGPT, and it put out better and more creative text than most of the GMs I've played with. It can be done with the right prompting.

[Rc2F6]

Ursprünglich geschrieben von Masterman397:

https://www.youtube.com/watch?v=Ba7pipuRfBs

The guy screwed these tools in a sandbox for kids called Modbox. So the technology is primitive and you can screw it anywhere and however you want. Moreover, you can download one of the GPT-3 or GPT-4 versions from GitHub for free.

https://youtu.be/jH-6-ZIgmKY

[Rc2F6]

Ursprünglich geschrieben von Voelkar:

Ursprünglich geschrieben von zelosdomingo:

You can just spin it as misdirection, if they've got something to hide. Why people who are supposed to be neutral or helping you are saying crazy stuff would be harder to explain... People do just say weird stuff sometimes in real life, I guess.

Still, the amount of falso information given by GPT-3 vastly outweighs the correct one. I guess it wokrs kinda the same in real life but not as often as GPT-3 does with comepletely made up information that it pulled out of thin air

If you look at two videos in the comments about this technology, you will see that you can strictly set the range of information from which the neural network will form grammatically formed phrases and with a semantic load. There will not even be a hint about false information, if you write to write false information from which the neural network will make up lies, then only in this case there will be lies. So you have very old information about this technology.

[Rc2F6]

Ursprünglich geschrieben von Tegiminis:

Please do not put a large language model into this game.

Beyond the cost and technical issues outlined above, the fact is that LLMs simply aren't very good at generating ~actual good text~. They are okay at recontextualizing rigid information into a conversational tone, and sometimes mimicking an existing writer, but that is their limit. They are novel precisely once, then they are irritating. I find they only impress people who aren't writers, who I should point out can instantly see through the generated text to the prompt.

I have my problems with how mechanistic the social experience in this game is, but slathering over it with an LLM will not fix any issues, and will in fact just make it more annoying and less consistent to interact with the characters.

Please hire a narrative designer instead. This game has a lot of promise, but it's still lacking in the writing department. Working with a human being aligned with your goals and with their own creative energy will do more for this game than jumbling conversations up with an LLM.

If you write two words into a neural network, it will form your boring phrases from them. And if you write a description of the world, it will make the dialogue cooler than any writer. You probably just aren't familiar with this technology at all, or you have old information. I advise you to update your information about neural networks. Your version will cost developers more. If you were familiar with this technology, you would see that this technology was used by a man in a children's sandbox called ModBox, created by him on his knees in 2 minutes and 24 seconds. And if you put in the effort and time for at least a week. Then the game will definitely sparkle with new colors. I wanted to ask if you are happy with the dialogues now? And so it turns out. And for a procedurally generated world, stupidly clogged dialogues as you suggest will be a huge minus. And the neural network, when properly fed, will provide endless dialog options for a procedurally generated world. Just what this project needs.

[Trentscousin]

We will have habitats on Mars before we have a full working LLM like GPT-4 in [this game].

To all those with weaksauce PC's, now imagine (if you can), taking this game, making it in Unreal engine 5.3, with full irremovable parallax ray tracing and 4D Shadows, with ultra bovine scrotum scrolling added.

I predict at 16K resolution, you will struggle to get 60 FPS.

Take this "idea" and put it with the "need guns, need MP".

With all due respect,

Trentscousin