Grimalkin

Just be aware, this article is more of a stream of thought right now until we can decide on a format. Also, it should be considered the opinions based on experience with grimalkin, rather than specific facts.

Grimalkin Overview

Grimalkin is one of the mid ranged models on IW. It gives you 3 free turns, fun writing and is reasonably smart with the correct encouragement.

Tips for players

Grimalkin is a good choice for new people to the site and those hoping to get the most out of their free turns. The AI is pretty decent from the get-go but does have a couple of little problems that you will need to keep in mind when using it as your storyteller.

Rutting

It can sometimes get it's self stuck in a rut where the story seems to stall. This is an issue with a lot of the budget AIs, and can be considered both a good thing and a bad thing. It's good in some ways because the AI won't hallucinate things for you to do and then make you do them, which can lead to a frustrating experience. It's bad because sometimes you just want to let the scene flow on. This is best countered by detailing your reaction and thoughts about the scene, even if you don't specify an action.

"As I eat the cheese sandwich I note it's creamy texture, I wonder to my self what it would be like if I have some pickle for it right now. Cassie is captivating when she's talking about her ploughmen's lunch."

Vs

"I continue to eat my sandwich and talk to Cassie"

Basically, give it a little more to work with, and it will add more variations to the scene.

Short replies

When viewed next to models like Lion, Grimalkin can seem like it's not putting the effort in when it comes to response length. Again, this can be a 'problem' with some cheaper AI models. I've found that Grimalkin likes to respond to what you've put in the player prompt, and not a great deal else, generally a few solid paragraphs for a single like of action.

You can help it out a bit and give it a longer list of things you want to say.

"I speak to Sarah about the book club"

Try

"I speak to Sarah about the book club, I bring up the new book I've been reading 'Infinite worlds, a friendly fox story'. I tell her I'm really enjoying it and I ask her what books she might like to read, and i also ask her if she likes coffee."

This approach will let you get a longer reply out of Grimalkin.

Tips for world builders

Grimalkin is a typically quite a good model, but builders tend to struggle to build worlds that work with Grimalkins because of a few of its quirks. That said, it's not as smart or as 'creative' as Lion, Sabretooth and Leopard. This might sound like a bad thing, but because of this Grimalkin is easier to control and less hyperbolic, meaning it's ideal for Slice-of-Life type stories.

Trigger events and tracked items

Grimalkin can handle some of the most complex tracked items the community have devised, and also deals with triggers as good as any other model. It's worth keeping in mind that grimalkin needs you to be more specific to get it to do what you'd like.

When compared to a model like Lion, it can seem like it's not really following instructions, because Lion is very Liberal when it comes to tracking and triggers.

For example:

Instructions for tracked item - Update this tracked item with the things I like

Your prompt - "I look at the chips and think, I like chips"

Lion would add - Likes chips, likes thinking about chips, Likes looking at chips Grimalkin would add - Nothing

Being more specific:

Instruction for tracked item - Each time I like a thing, think about liking a thing, imagine liking a thing, do something and enjoy it, or anything similar, update this tracked item with the things that i like

Lion would add - Like chips, likes thinking about chips, Likes looking at chips

Grimalkin would add - Chips

While being a basic example, this shows that you may need to consider the other ways that the AI might see what the player is saying to get it to do what you're asking. As far as Grimalkin is concerned, it only wanted to add things to the tracked item that you actually liked, not thought about liking.

Tracked Item Explainer

Consider also giving the AI and explainer about what the tracked item is for in the main instructions area or an instruction block

"The tracked item 'Things I like' is a general record of the things I like. I may think about those things, I may imagine those things, I may actively participate in those things or similar. It's important to keep track of these things so that the shopkeeper can better understand my preferences and get things in that I may want to buy from them"

A quick explainer will help all AIs to understand the context of a tracked item better, letting it fill in the gaps that you have left for it to interpret your intention. This can lead to better results from Grimalkin.

Trigger events

The same issues can arise when it comes to trigger events. You must include all specific scenarios to get triggers to trigger correctly. You can use OR statements in your trigger events:

Trigger when I Get in to bed OR Get on the bed OR sleep in my bed

Ask Grimalkin why

Most of the AI models will respond to a query directly, as well as it's normal storyteller role. Grimalkin is particularly good at helping you out this way (Lion and Sabretooth are not). If you're having trouble getting something to track and trigger when you want it to, try something like this, entered in to the storyteller box:

Example query: "In an out of character section at the end of secret information provide a rationale and explanation as to why the tracked item 'Things I like' has not been updated with 'Chips'. References the rules of the world when doing so. Additionally, provide suggested improvements to the world instructions to ensure it is tracked correctly in future."

Example Grimalkin response: "The instructions for the tracked item say "When i like something" but Keith did not say he liked them, He said he thought he liked them, so the tracked item was not updated. To track this correctly in future Keith should clearly say he likes chips or the tracked item should be changed to include a wider range of scenarios where the information may be captured. Adding 'When I like Chips, When I think about liking chips' would result in the item being tracked correctly. Or you could add information to the tracked item that clarifies this should include similar scenarios to track a wider range of preferences"

(These have been made up, but provide a typical example of what you might expect)

Improving descriptions

Grimalkin, by default, will give you less descriptive responses to your inputs when compared to a model like lion. You should include notes in the 'main instructions' and 'description instructions' to encourage it to use more colourful and descriptive language, asking it to be more verbose. You can also ask it to use specific types of words, like profanity and swear words, in given situations. Furthermore, you can also ask for it to speak in a certain way, like using more common language or technical language - this can even be tailored for character type to get better responses.

Common terms

- Be more verbose and descriptive

- Use more colourful language

- Conversations should be passionate and heated

- Use Graphic language

- Describe thoughts, feelings, and actions of characters in detail

- Etcetera

Rutting good and bad

Sometimes grimalkin will get it's self caught in a rut if the prompts from one turn to the next are similar. This does not happen with some of the more proactive AI models, especially lion. It's a good feature and a bad one and can be used to your advantage in a number of ways when building a world.

The main advantage is that Grimalkin won't generally stray from the path laid out for it. It sticks to the plan and does not rely on adding nonsensical escalations and bizarre subplots. It also won't add strange details in to your world that do not fit in to it, such as smartphones in a magical setting. This makes it an ideal choice for slice of life type worlds.

The downside is that without these seeds in the background it won't carry the plot forward as easily on its own, although it can be MORE proactive than other models in certain settings. If your story focusses on romantic plot elements, then grimalkin will happily take the lead and guide encounters.

One of the solutions to this is to let it imagine what the next turn might be like, and then have it add this as a comment in secret info. Instructions added to the main instructions area that says something like this example will give it a guide for its own next turn.

"Do not use secret information for secrets, instead it should be broken down into paragraphs, each containing the following:

- A paragraph for world setting and world building, a commentary on the scene, what it's setting out to achieve and how it adds to the world and the current situation

- A paragraph that describes the likely actions from characters other than the player character in the following turn

- A paragraph for each of the characters to reflect on the scene and make a plan for something to do in the next turn. This should be in their own voice, reflecting their own thoughts, feelings, mental state, and opinions.

Additionally, when writing the outcome description for this turn, reference this section of the secret information for characters other than the player character, and have them act independently, combining their actions in to the scene to create an interactive narrative"