Install Steam
login
|
language
简体中文 (Simplified Chinese)
繁體中文 (Traditional Chinese)
日本語 (Japanese)
한국어 (Korean)
ไทย (Thai)
Български (Bulgarian)
Čeština (Czech)
Dansk (Danish)
Deutsch (German)
Español - España (Spanish - Spain)
Español - Latinoamérica (Spanish - Latin America)
Ελληνικά (Greek)
Français (French)
Italiano (Italian)
Bahasa Indonesia (Indonesian)
Magyar (Hungarian)
Nederlands (Dutch)
Norsk (Norwegian)
Polski (Polish)
Português (Portuguese - Portugal)
Português - Brasil (Portuguese - Brazil)
Română (Romanian)
Русский (Russian)
Suomi (Finnish)
Svenska (Swedish)
Türkçe (Turkish)
Tiếng Việt (Vietnamese)
Українська (Ukrainian)
Report a translation problem
Give it a try after the next update. I'll try to think of a different way to add information to the bots.
I second this. If someones using a free model and doesn't care about token usage, automatically prompting the nearest bot with your chat message would feel alot more natural. Even if you still had to include a !.
Example: "#how are you doing today?" The message would then prompt the closest bot to your character.
You would almost need a second decision-based AI just to translate their sentence into an in-game command.
For example having an AI give an angry response: "I'm going to get you for that! Your out of line!"
and then a second decision AI to decipher it to "aggressive" and then use the decision AIs response to apply it to an in-game bot behavior/command.
I know this type of decision AI to group strings into specific categories exist because I was looking at one on github a while back for possible future use.
This would double the API calls and increase the token use, but I think it will work. I am not entirely sure how accurate it will be, but maybe having the bots misunderstand their orders sometimes will be a funny quirk.
If you can get the direction the player is currently facing you can use that. Naturally you would want to get the closest person your character is facing towards.
You could just include a max distance before using the closest person behind you.
Or get the players cursor position and use that instead of their character position.
So players input would be: "John, get to the guns!" - Order to turret position.
Bots input: "Sorry captain but I cant do that. You'll need to get a professional." - Said no.
Result: Ignore.
Token cost isn't a concern if there are any free models that can perform categorization. But it may have to be simplified to only accounting for the players command to the bot and bot responses by themselves.
Just tested this theory with openchat/openchat-7b:free model and it has been able to successfully categorize orders along with whether they were obeyed or not.
Here were my prompts: "I need you to categorize any responses into the following categories: order to guns obeyed, order to attack obeyed, angry and refused, ignored."
My response: ""John, get on the guns!" "No thanks captain. You'll need somebody else for that.""
Decision AI response: ""John, get on the guns!" "No thanks captain. You'll need somebody else for that." - Angry and refused"
Search the return string for "Angry and refused" and you could translate this into the bot ignoring your commands in game.
So you say "John, get on the guns!". I send that to AI to determine what type of order it is, it determines it's asking John to get on a gun, and then I assign the order to the bot.
A few moments later, I check if getting on the guns is an active order for the bot. If so, I include something in the prompt about the bot accepting the order.
If the bot rejected the order, he would no longer have that as an active order, so I could include something in the prompt about him rejecting the order.
It would probably be easier on tokens and the coding side if you ran both the players text to the bot and the bots response at the same time.
Example prompt to the decision AI: "Players text sent to bot" "Bots response"
And then it could decide whether to set a command for the bot or not. If the AI decided the bot ignored the order it simply never gives the command to go to the guns.