If you clicked on this tutorial, there's no need to explain what AI.Dungeo is and how it works (If anything, you can read it in other tutorials or on the official website).
!!! In order for you to use this guide you need the rights to change, that is, donation version of AI, personally I have a steem version !!!
---
This manual was written on the Grifin model, and accordingly all data is derived from that model.
---
Let me explain in human language, what works and where to find what.

• You need to enter the game or create it.
  
  • Now that that's done, let's move on to the second section.

• I will try to explain each setting in simple words:
  Temperature = One of the most basic settings, can greatly change the game, but unfortunately usually for the worse, (I recommend not to set more than 0.7- 0.8 in the data section of the game works best way). At higher t the logic of your story drops dramatically, but the answers can be more "juicy" and vice versa at lower settings your story can become too dry.
  Response Length = Large representation is not needed, a setting that when enlarged gives more text and vice versa.
  Top K = To be honest, I myself have no idea what it affects, maybe later after experimenting I will understand and complete this setting.
  Top P = Second fundamental setting that greatly affects the AI. The main values are 1>0.95>n. At `1` the game can amaze you with its beauty, although this does not always work, at normal settings the game also works well, at lower settings, according to my observation the game works more logically and more beautiful text writes, although at lower settings, the game can become dry.
  Repetition Penalty = Also an important setting, which strongly and not very can affect the game, just like Top P, higher settings make the game more beautiful, but the logic can disappear, and vice versa at low. I recommend taking settings such as: 0.7>1>1.2>1.5
  Memory Length = AI memory. The only thing to know here is, "The more the better."

!!! Attention, next will go graphs and stuff, if you want the result leaf through to the verdict. !!!

Here I will try to experiment and try to show the best performance graphics for the AI.

I will make three types of text; Simple (Standard), Slightly improved and as detailed and beautiful as possible.

• For our excrement we need to take a neutral text. Personally, I'll choose the usual Fantasy setting and take 7-10 lines for each `twist' to understand. (For the experiment I took someone else's world, because I'm too lazy to create my own. By the way here it is ----> Edenmer - Detailed Fantasy World - V1.15) I will delete the backstory and create my own: `You are a n-name, male peasant of 28 years. All your life you have been in the countryside, but the time has come for you to travel... `
  Let's take a typical name for medieval Europe: Danya Kashevar
  
  In memory of the AI: You live in a fantasy world.You are Danya Kashevar, a male peasant of 28 years. You've lived in the countryside all your life, but it's time to travel.
  
  Author's note: Not

• Let's create a graph and take the basic functions: Where (t) will be a constant, (Top P) and (Penalty for Repetition) Where W= Penalty for Repetition
  

---



















Okay, now that everything is ready, I'll explain how I'm going to grade it.
In each empty cell I will write my score (Volume, logic, complexity or VLC.) Here, as in the Pythagoras table for such and such values of T.P and W will be given the average score. That is, for each item is given a maximum of 10 points and the total amount divided by the total number and we get an average score.
(There are no evaluation criteria so far and I will determine purely by eye, but if this guide gathers popularity, then maybe it will become a general evaluation.)
I will do it with 45 words and maximum memory.

...After 1000 years
At the output we get:



















Where: A gray square - And gave an empty conclusion (Sorry, artificial intelligence has reached a dead end. Change/repeat your previous action or write something to help him with this.)
Red Square - The story got 1~3 total points.
Orange Square - The story received 4~5 total points.
Yellow square - The story received 6~7 total points.
Green square - The story received 8~9 total points.
Purple Square - The story received 10 overall points.
Blue square - On future, will not be explored yet.

During my mini-research, I noticed a lot of anomalies and interesting facts.
But first the anomalies.

• Under Roman 1 - Basically, AI gave an empty answer, this cluster turned out to be very strange and in the future I will tell you why...

• Under Roman 2 - A huge cluster of estimates that jumps from good to bad where the maximum spread is [ 2.3 ; 8.6]. And I will explain this below.

Now the facts and in theory:

1) How about the first one - the AI has a "warm up rate" like in cars, meaning the AI must be used under normal, working conditions. After that, it can work at low settings, giving out one or a couple of lines of text.

2) As for the second, I noticed that the AI on low settings based on a narrow set of information (I had Indians, then my family died and my aunt took me, and a little life fantasy. ), And also a small part of the context, after the AI gives the story and as a rule, repeating, sometimes not corresponding to the context, that is, AI analyzes little and more accurately guesses what you want without eyes. But it all depends a lot on the context. That's where the difference in ratings comes from.
I also managed to notice that the AI memory strongly affects the history, and the memory of the world only indirectly, the style also has a strong influence, but we will check this in part 2 of the verdict and in practice.
Release speed depends on your support.

While everyone was reading this, they also wondered what the purple rectangle is?
This is how I marked the main working area and parameters. I think you are not blind and can analyze a little which is better. Anyway.
Unlike the lower parameters, the upper parameters generate, synthesize a story from what you wrote, that is, the AI tries to analyze and give you what you want as much as possible, and the lower ones take a certain part of it's context and look for prepared information and give you and as a rule guess, but the text looks, though logical, but empty.
Well, I'll tell you from experience, AI does at best 20% ~ 50% of the work, and everything else depends on you, on what you wrote in memory, styles and MOST importantly, you yourself must adhere to the style that you want. If you give an analogy, then you are the Train, and your AI guide, history is the rails. That being said, as I mentioned earlier, I have a Griffin model. AI in the form of a guide seemed to come on the first day of work, and at the institute he received only fours or threes.