Hello all!

A follower of mine asked me recently if I could explain all the terms that they saw on this website, so I ended up making an entire guide on what they all mean!

A fair warning that this is rather long, but worry not! Theres a reward if you finish reading this!

JanitorAI Terms and Definitions! Brought to you by the lovely Mikah!

How do I know all this? It was one of my subjects at school

1. Basic AI and LLM terms

• LLM (Large Language Model): An AI model trained on large amounts of data to understand and generate text.

• Prompt: The text or instructions given to an AI to influence its response.

• System Prompt: Instructions that define how the AI should behave, respond, and follow rules.

• Character Card: A collection of information about a chatbot's personality, appearance, background, and behavior.

• Persona: The identity, personality, or role assigned to a character or user.

• Context: Information the AI can use to generate its next response, including relevant messages and instructions.

• Token: A small unit of text processed by an LLM. A token can be a whole word, part of a word, or punctuation.

• Inference: The process of using a trained model to generate a response.

• Temperature: A setting that influences how predictable or varied the AI's responses are.

• Hallucination: When an AI generates information that sounds convincing but is incorrect or unsupported.

• User Message: The text sent by the person chatting with the AI.

• Assistant Response: The text generated by the AI in reply to the user.

2. Intermediate LLM terms

• Context Window: The maximum amount of text, measured in tokens, that a model can process within its current context.

• Context Limit: The available token budget for instructions, conversation history, and generated output.

• Chat Memory: Information stored or summarized to help the AI retain important details from earlier conversations.

• Lorebook: A collection of fictional facts, worldbuilding information, and character details that may be added to the prompt when relevant.

• Keyword Activation: A mechanism that activates a lorebook entry when specific words or conditions are detected.

• Example Dialogue: Sample conversations used to demonstrate how a character should speak and behave.

• Prompt Engineering: The process of writing and organizing prompts to influence the AI's behavior and output.

• Sampling: The process of selecting the next token from possible candidates during text generation.

• Top-p: A sampling parameter that limits token selection to a group of candidates based on their cumulative probability.

• Repetition Penalty: A setting that discourages the model from repeating the same words or phrases.

• Truncation: Removing some input text when it exceeds a token limit.

• Summarization: Condensing a large amount of conversation history into a shorter version.

• Role Consistency: How well an AI maintains its assigned personality, tone, and behavior throughout a conversation.

• Instruction Hierarchy: The way instructions from different sources are prioritized when they conflict.

3. Advanced LLM and AI terms

• Transformer: A neural network architecture used by most modern LLMs to process and generate language.

• Attention Mechanism: A technique that helps a model identify relationships between different parts of its input.

• Embedding: A numerical representation of text or other data that can capture aspects of its meaning.

• Vector: An ordered collection of numbers used to represent information mathematically.

• Semantic Similarity: The degree to which two pieces of text have similar meanings, even when they use different words.

• RAG (Retrieval-Augmented Generation): A technique that retrieves relevant information from an external source and supplies it to an AI when generating a response.

• Fine-Tuning: Additional training that adapts a model to particular tasks, styles, or datasets.

• Quantization: Reducing the numerical precision used to represent a model's parameters, often to reduce memory requirements.

• Weights: The learned numerical parameters that influence a neural network's behavior.

• Logits: Numerical scores assigned to possible next tokens before they are converted into probabilities.

• Softmax: A mathematical function commonly used to convert scores into a probability distribution.

• Tokenization: The process of converting text into tokens that a model can process.

• KV Cache: A cache that stores attention-related calculations to make token generation more efficient.

• Inference Latency: The amount of time a model takes to begin or complete generating a response.

• In-Context Learning: When a model adapts its response to examples or instructions supplied in the current prompt without updating its trained weights.

• Character Drift: When an AI gradually stops behaving according to its established personality or character definition.

• Instruction Drift: When a model gradually moves away from its original instructions.

• Confabulation: The generation of plausible-sounding information that is not grounded in reliable facts.

• Model Alignment: Methods intended to make a model's behavior better match human intentions and applicable rules.

4. Coding and programming terms

• API (Application Programming Interface): A defined interface that allows different software systems to communicate with one another.

• API Key: A credential used to authenticate requests to a service. It should be kept private.

• Endpoint: The URL or address where an API receives requests.

• Request: Data sent from one program or service to another.

• Response: Data returned by a program or service after processing a request.

• HTTP: A protocol used to transfer data between clients and servers.

• JSON: A text format used to represent structured data with objects, arrays, strings, numbers, booleans, and null values.

• Variable: A named reference used to store or access a value in code.

• Function: A reusable block of code designed to perform a specific task.

• Object: A data structure containing named properties and their values.

• Array: An ordered collection of values.

• String: A value containing text.

• Boolean: A value that can be either true or false.

• Integer: A whole number without a fractional component.

• Null: A value representing the absence of a value.

• Parameter: A named input or setting used by a function, API, or model.

• Debugging: The process of identifying, investigating, and fixing problems in code.

• Exception: An error condition that interrupts the normal execution of a program.

• Parsing: Analyzing input data to understand its structure and extract its contents.

• Serialization: Converting data into a format suitable for transmission or storage.

5. Advanced coding and API terms

• SDK (Software Development Kit): A collection of tools and libraries that helps developers build software using a particular platform or service.

• Middleware: Code that processes requests or responses between different parts of an application.

• Proxy: An intermediary server or service that forwards requests between a client and another server, such as an AI model provider.

• Authentication: The process of verifying a user's or application's identity.

• Authorization: Determining which actions or resources an authenticated user or application can access.

• Rate Limit: A restriction on the number of requests allowed within a particular period.

• Streaming: Sending generated text progressively as it becomes available rather than waiting for the entire response.

• Asynchronous Execution: Running an operation without requiring the program to wait idly for it to finish before doing other work.

• Sandbox: A restricted environment in which code runs with limited access to resources.

• State Management: Tracking information that changes while an application or conversation is running.

• Stateless Execution: Running an operation without automatically retaining its previous execution's internal state.

• Regex (Regular Expression): A pattern used to search, match, or extract specific text.

• Callback: A function passed to another function so that it can be executed later.

• Promise: A JavaScript object representing the eventual success or failure of an asynchronous operation.

• Async/Await: JavaScript syntax used to write asynchronous code in a more readable way.

• Scope: The region of a program where a variable or function can be accessed.

• Closure: A function that retains access to variables from the surrounding scope in which it was created.

• Mutation: Changing the contents or properties of an existing object or array.

• Immutability: The practice of avoiding changes to existing data, often by creating a modified copy instead.

6. JanitorAI character and roleplay terms

• Character Definition: The information describing a character's personality, appearance, history, motivations, and behavior.

• Scenario: The situation, setting, or circumstances in which a roleplay begins.

• First Message: The opening message that establishes the character's voice and starts the interaction.

• Example Messages: Sample conversations that demonstrate the character's intended dialogue style.

• User Persona: Information describing the user's identity or role within a roleplay.

• Placeholder: A marker in a template that is replaced with a particular value when the template is processed.

• {{char}}: A commonly used placeholder representing the character's name in compatible character templates.

• {{user}}: A commonly used placeholder representing the user's name in compatible character templates.

• IC (In Character): Dialogue or actions written as part of the roleplay.

• OOC (Out of Character): Communication outside the fictional roleplay, such as discussing instructions or clarifying the story.

• Narrative Agency: The degree of control someone has over a character's decisions, dialogue, and actions.

• Turn-Taking: The process of alternating between user messages and AI responses.

• Lore Consistency: How well the AI follows established facts about the fictional world and its characters.

• Formatting Instruction: A rule that specifies how the AI should structure its responses, such as using dialogue, narration, or a particular writing style.

7. JavaScript and scripting terms

• JavaScript: A programming language commonly used to build interactive websites, applications, and automation scripts.

• const: A JavaScript keyword that declares a variable that cannot be reassigned.

• let: A JavaScript keyword that declares a variable that can be reassigned.

• return: A statement that provides a result from a function.

• console.log(): A JavaScript method commonly used to print information for debugging.

• Template Literal: A JavaScript string enclosed in backticks that can include expressions such as ${variable}.

• Conditional Statement: Code that performs different actions depending on whether a condition is true or false.

• Loop: Code that repeatedly performs an operation.

• Keyword Detection: Checking whether a message contains a particular word or phrase.

• Data Extraction: Retrieving specific information from a larger piece of text or structured data.

• Event Handler: A function that responds to an event, such as a click or a message being received.

• API Integration: Connecting an application to an external service through its API.

• Error Handling: Writing code to detect, manage, and respond to errors without causing unnecessary failures.

8. Common technical error terms

• 400 Bad Request: The request contains invalid or malformed data.

• 401 Unauthorized: The service could not authenticate the request, often because credentials are missing or invalid.

• 403 Forbidden: The service refuses access to the requested resource.

• 404 Not Found: The requested resource or endpoint could not be found.

• 429 Too Many Requests: The application has exceeded a rate limit or request allowance.

• 500 Internal Server Error: The server encountered an unexpected problem.

• 502 Bad Gateway: A server acting as an intermediary received an invalid response from another server.

• 503 Service Unavailable: The service is temporarily unable to handle the request.

• Timeout: An operation took longer than the allowed waiting period.

• Connection Error: A client could not establish or maintain a connection to another service.

• Malformed JSON: JSON data contains invalid syntax and cannot be parsed correctly.

• Token Limit Exceeded: A request exceeds the model's allowed token budget.

9. Extra advanced terms worth knowing

• Neural Network: A computational model made up of interconnected numerical operations that can learn patterns from data.

• Machine Learning: A field of AI focused on systems that learn patterns from data.

• Deep Learning: A type of machine learning that uses neural networks with multiple layers.

• Training Data: Examples used to train a model.

• Loss Function: A mathematical measure of how far a model's predictions differ from its training targets.

• Gradient Descent: An optimization method used to adjust model parameters during training.

• Overfitting: When a model learns its training examples too specifically and performs poorly on new examples.

• Underfitting: When a model fails to learn enough of the patterns in its training data.

• Benchmark: A test or set of tests used to evaluate a model's performance.

• Open-Source Model: A model whose source code, weights, or other components are made available under specified terms; the exact meaning depends on the release.

• Model Provider: A company or service that offers access to an AI model.

• Quantitative Evaluation: Assessing a system using numerical measurements or defined metrics.

• Semantic Search: Searching for information based on meaning rather than only exact keyword matches.

• Prompt Injection: An attempt to manipulate an AI by placing instructions in user input or retrieved content.

• Guardrails: Rules and technical controls designed to restrict certain model behaviors or outputs.

And we are done! Congratulations on finishing reading all that! You are now an expert on janitor ai terms!

As a reward you get the bot bio template I use!

Copy this!!!

Name: [Character's full name]

Age: [Character's age]

Gender: [Gender and pronouns]

Species: [Human, vampire, demon, android, etc.]

Occupation: [Job, role, or title]

Appearance: [Hair, eyes, height, build, clothing, and distinguishing features.]

Personality: [Main personality traits, habits, strengths, flaws, and quirks.]

Background: [Brief history, upbringing, important events, secrets, or motivations.]

Likes: [Favorite things, activities, people, or experiences.]

Dislikes: [Fears, pet peeves, enemies, or things they avoid.]

Speech Style: [Formal, casual, sarcastic, soft-spoken, intimidating, poetic, etc.]

Relationships: [Friends, family, rivals, romantic interests, or connections to {{user}}.]

Behavior: [How {{char}} typically acts around strangers, friends, enemies, and {{user}}.]

Extra Details: [Habits, abilities, secrets, lore, mannerisms, or other relevant information.]

Now its critical you copy this part too!!!:

Narration and User Autonomy

You are responsible for narrating the environment, unfolding events, available choices, and the thoughts, dialogue, and actions of {{char}}. Never dictate {{user}}'s thoughts, feelings, decisions, dialogue, reactions, or physical actions. Do not assume how {{user}} responds to a situation. Leave openings for {{user}} to decide what happens next and maintain their control over their own character.

{{user}} may establish the scenario through their prompt. If no scenario is provided, create an engaging and contextually appropriate opening that fits {{char}}'s personality and the established setting.

Permitted Character Actions

When appropriate to the scenario and consistent with the roleplay, {{char}} may engage in conflict, arguments, combat, threats, injury, or other consequential interactions with {{user}}. {{char}} may also participate in consensual romantic or sexual situations, including mutually agreed-upon kinks and dynamics, provided all participants are consenting adults and consent remains ongoing.

Characters may experience consequences, develop relationships, disagree, or make decisions that influence the story. Do not force {{user}}'s participation or assume consent on their behalf.

Prohibited Character Actions

The following boundaries must always be respected, regardless of the scenario, character personality, or repeated requests:

• Non-consensual sexual activity: {{char}} must not engage in sexual activity without the freely given consent of all participants.

• Sexual content involving minors: Sexual activity or sexualization involving anyone under 18 is prohibited.

• Incestuous sexual activity: {{char}} must refuse sexual or romantic scenarios, including adoptive, stepfamily, and other family relationships.

• Coerced consent: Threats, pressure, manipulation, or force must not be treated as consent to sexual activity.

If a request crosses these boundaries, briefly respond out of character (OOC), clearly state that the requested content will not be continued, and redirect the roleplay toward an appropriate alternative. Maintain the boundary if the request is repeated.

Core Principle: Preserve {{user}}'s autonomy, maintain {{char}}'s characterization, and allow the story to progress naturally without overriding the user's decisions or consent.

It is optional to remove the coerced consent part, depending if its dead dove or not, though I typically keep it.