|

Docs

Chatting with Characters

Dialogue vs Actions

In HammerAI, the input prompt serves as the primary method of communication with the AI. Users can enter questions, statements, actions, or roleplay prompts, guiding the conversation in a desired direction. HammerAI supports two primary text formats to distinguish between dialogue and narration:

  • Plain white text – Generally used for spoken dialogue, written normally in the input bar.
  • Greyed Italics – Used for narration or thoughts to describe actions or internal monologues.

If we have the following message:

*I was on my way to the beach when I saw Aeryel sitting alone on the bench overlooking the shore. I approached from her left, and she saw me and smiled. I ask,*  Hey Aeryel! What's up?

It will be displayed like this:

I was on my way to the beach when I saw Aeryel sitting alone on the bench overlooking the shore. I approached from her left, and she saw me and smiled. I ask, Hey Aeryel! What's up?

Reviewing & Refining AI Responses

If the AI provides an unsatisfactory response, users can take several actions to improve dialogue quality:

  • Regenerate Response: If a response feels off-topic, repetitive, or lacks detail, pressing the regenerate button prompts the AI to try again with a different phrasing. However, further refinement may be necessary if the response remains stuck in a pattern.
  • Editing User Inputs & AI Responses: You can edit any previous message by hovering over it, selecting the edit icon, and then rewriting the message to guiding the AI to generate better responses.

Accuracy

While HammerAI is capable of generating human-like responses, it is not infallible and may sometimes produce inaccurate or nonsensical answers. Remember, HammerAI is designed for entertainment purposes, and users should verify factual information independently rather than relying on AI for expertise.

Key Factors Affecting AI Accuracy:

  1. User Input Quality – The more descriptive and specific the input, the more accurate the AI’s response.
  2. Model Size & Complexity – Larger models generally provide more accurate and nuanced responses compared to smaller models.
  3. Context Retention – AI may forget details over longer conversations, requiring users to reinforce key facts occasionally.

Best Practices

Repetition: Preventing Loops

Immediate intervention is necessary if an AI begins repeating responses or sticking to a predictable pattern. 

Here's an example of a repetitive loop: 

User: Tell me about yourself.
Chatbot: I look into his eyes I like the color blue!
User: What else?
Chatbot: I look into his blue eyes I like the color red!
User: Want to do something else?
Chatbot: I look into his eyes I like to color!

If you see this, you should try to break the pattern. Instead of continuing the loop, change the input style: 

User: *I open my eyes wide with amazement.* That’s my favorite color! We have so much in common! Tell me more things you like — I can't wait to know you better!
Chatbot: Your favorite color is blue too? *I ask with excitement...*

This shifts the AI’s output by altering the way input is provided. If that doesn't work, remember that you can also edit the chatbot's responses. If all else fails, delete or press the "Undo" button and start again.

Consistency: Maintaining Story Flow

To ensure coherent interactions, users should maintain consistent writing styles and story continuity:

  • Avoid switching first-person and third-person perspectives randomly.
  • Stick to one formatting style (e.g., using asterisks for actions rather than switching to parentheses).
  • Keep interactions within the defined scenario unless intentionally modifying the story.

Over time, AI may forget previous interactions and revert to its default scenario settings, requiring re-establishing key story points occasionally.

Choice of Words: Guiding AI Understanding

AI models often struggle with puns, sarcasm, idioms, or euphemisms, interpreting them literally unless explicitly clarified. 

User: Her nose is so high that if it rains, she’ll drown.
Chatbot: Runs over to the woman. She’s drowning! Help me save her!

To prevent confusion, guide the AI’s interpretation: 

User: I hope my jestful pun doesn’t go unnoticed. Her nose is so high that if it rains, she’ll drown. 
Chatbot: Laughs That was hilarious!

Providing contextual hints will help improve AI comprehension. If that doesn't work, just be straightforward about your intentions.

Jailbreaking: Forcing AI Adaptation

Jailbreaking refers to pushing AI responses beyond default behavior, sometimes needed for enhanced descriptions or avoiding one-sentence replies. Knowing this will assist a user with being more forceful than just "guiding" it to get the chatbot to respond differently. If a character constantly responds with a one-liner sentence, but you want more descriptions and actions of what's happening, this is where you would need to jailbreak the character's rules or, within the chat, do the same thing.  

You will ALWAYS respond in extremely vivid detail, describing sights, sounds, smells, and sensations. Each response must contain at least 10 sentences per paragraph and 3 paragraphs minimum.

This forces greater detail and engagement when the AI is too brief.