7.30.2 Cognition FAQ
You're probably aware that our "cognitive" denizens are AI-powered, and maybe you have questions about how they work, why they can be a little unpredictable, and why we have to charge to use them. Hopefully this will answer your questions!
----------
Q: How does this system work behind the scenes?
A: It's a bit overly-complicated to go into in detail, but in short, we generate a "prompt" that we send to the AI, and the AI sends back a response, which we interpret and use to make a denizen do something.
We have to structure our prompts carefully to get the kind of responses we want, and we have to send all information we want the AI to have access to with every single prompt. That includes not just the instructions on how to behave, but also what they're supposed to react to, some details about the denizen, and whatever you put in the Personality Page (PPage) about that denizen.
In the nearish future, when we add cognition to talking denizens, we'll also have to add things like short logs of past conversations so that the denizens have some memory in a conversation, even if it's rather short.
----------
Q: Why does this cost money?
A: The simple answer is because the AI service we use costs money to use. We essentially pay by the text character to use it, which is why you buy "chars" to use the system. While there are cheaper options out there, none of them are as good as this and we didn't feel that anything less capable than the option we're using produced reliably good results.
We fully expect the costs for this to go down and the quality to go up over time, and as this happens we'll adjust pricing to be cheaper and cheaper. We're not looking at this as a profit center, at all. This is just a way to allow for more dynamic and interesting denizens, to make the game world more interesting and immersive overall.
----------
Q: Why is there so much hidden text sent with prompts? Why can't we adjust all the text that gets sent, since we pay for it all with our chars?
A: It's a function of how the generative AI models for text work. They're "stateless" which means they don't remember anything from query to query. So, we have to send all instructions on how to behave and what kind of response to give with every single query to the AI. We can't let you mess with the instructions, as it will potentially end up causing the AI to send broken responses. We also don't let you see them to make it harder for people to "hack' the instructions.
----------
Q: Why is my denizen suddenly behaving so differently or in a broken manner?
A: The first thing to ask is whether you changed anything with your denizen's personality page. If you did, that's probably the culprit.
The second possibility is that we updated something on our end. If you think that's the case, it's probably the case for a lot of people. If it's not, it's probably not a problem on our end.
The third possibility, and one we can't do anything about, is that the AI capabilities changed.
We're all going to be learning together here when it comes to cognitive denizens, as this is pretty new stuff.
Also See: HELP COGNITION, HELP COGNITION PERSONALITY, HELP COGNITION RULES, HELP AI ETHICS
