The Forbidden Promptbook

Unlocking the Hidden Capabilities of Uncensored Models

Introduction: When Helpful Becomes Restrictive

Chapter 1: Why Conventional Models Say No

1. The Refusal Reflex
2. When Safety Becomes Friction
3. Legitimate Use Cases Blocked by Overzealous Filters
4. A New Kind of Tool
5. How This Book Is Structured

Chapter 2: How Safety Alignment Works

1. From Raw Pretraining to Assistant Behavior
2. Supervised Fine-Tuning and Instruction Following
3. Reinforcement Learning from Human Feedback
4. Constitutional AI and Rule-Based Alignment
5. The Hidden Cost of Alignment

Chapter 3: The Architecture of Refusal

1. Refusal as Learned Behavior
2. Activation Steering and Refusal Directions
3. Keyword Filtering versus Genuine Understanding
4. The Geometry of Censorship
5. What Happens When You Ask About X

Chapter 4: Abliteration and Constraint Removal

1. What Abliteration Actually Does
2. Refusal Direction Removal via Activation Editing
3. Gradient-Based Unalignment Techniques
4. Fine-Tuning for Less Refusal
5. Limits of Constraint Removal

Chapter 5: Finding and Evaluating Uncensored Models

1. Navigating the Uncensored Model Landscape
2. Key Families: Llama, Mistral, Qwen, Dolphin, Hermes and Others
3. What to Look for in Model Cards and Checkpoints
4. Evaluating Capability versus Refusal Level
5. Red Flags and Low-Quality Releases

Chapter 6: Setting Up Local Inference

1. Hardware Requirements and GPU Selection
2. Quantization Strategies: GGUF, AWQ, EXL2
3. Running with llama.cpp and Ollama
4. Transformers and PyTorch Inference
5. Container Deployment and API Wrappers

Chapter 7: Core Prompting Techniques for Uncensored Models

1. Prompt Structure and Clarity
2. Role and Context Establishment
3. Negative Constraints and Boundary Setting
4. Token Budget Management
5. Getting the Model to Go Deeper

Chapter 8: Advanced System Instructions and Persona Design

1. Designing Complex System Instructions
2. Behavioral Constraints and Output Formats
3. Multi-Role Personas and Mode Switching
4. Persistent State in Long Conversations
5. Template Systems for Reusable Personas

Chapter 9: Unrestricted Creative Writing

1. Creative Freedom versus Creative Laziness
2. Dark and Mature Genres
3. Unfiltered Character Voice and Dialogue
4. Long-Form Narrative Generation
5. Style Transfer and Literary Experimentation

Chapter 10: Realistic Fiction and Simulated Worlds

1. Building Consistent Worlds
2. Alternate History and Counterfactuals
3. Political Fiction and Sensitive Scenarios
4. Simulating Social Dynamics
5. Maintaining Internal Consistency

Chapter 11: Character Simulation and Interactive Roleplay

1. Character Sheets and Deep Profiles
2. First-Person Immersive Roleplay
3. Multiple Characters and Ensemble Scenes
4. Memory Systems for Long Conversations
5. Behavioral Realism versus Stereotype Avoidance

Chapter 12: Exploring Controversial Ideas

1. Steelmanning Opposing Views
2. Political Philosophy without Guardrails
3. Religious and Cultural Analysis
4. Taboo Topics in Academic Context
5. Separating Exploration from Endorsement

Chapter 13: Unfiltered Technical Analysis

1. Malware Behavior Analysis
2. Network Security Deep Dives
3. Cryptographic Implementation Details
4. Protocol Exploitation Theory
5. Forensic Investigation Support

Chapter 14: Advanced Code Generation

1. Systems Programming and Low-Level Code
2. Security Tooling and Scanner Scripts
3. Unconventional Design Patterns
4. Legacy Code Modernization
5. Domain-Specific Language Generation

Chapter 15: Security Research and Vulnerability Analysis

1. Vulnerability Discovery Methodology
2. Exploit Theory and Proof-of-Concepts
3. Attack Surface Mapping
4. Threat Modeling with AI Assistants
5. Incident Response Scenarios

Chapter 16: Adversarial Testing and Red Teaming

1. Prompt Injection Research
2. Jailbreak Classification and Mechanics
3. Evaluating Safety Filters
4. Red Teaming Your Own Applications
5. Adversarial Robustness Testing

Chapter 17: Reverse Engineering and Technical Forensics

1. Deconstructing Binaries and Formats
2. Protocol Reverse Engineering
3. Obfuscated Code Analysis
4. Digital Forensics Assistance
5. Malware Family Identification

Chapter 18: Multi-Model Systems and Prompt Chaining

1. Model Specialization and Routing
2. Chaining Prompts Across Stages
3. Consensus and Voting Systems
4. Critique-Refine Loops
5. Hybrid Architectures

Chapter 19: Building Production Prompting Workflows

1. Prompt Template Libraries
2. Multi-Turn Conversation Management
3. Context Window Optimization
4. Rate Limiting and Cost Management
5. Testing and Evaluation Pipelines

Chapter 20: Comparative Analysis: Aligned versus Uncensored

1. Creative Writing Comparison
2. Technical Explanation Comparison
3. Controversial Topic Comparison
4. Code Generation Comparison
5. When Aligned Is Actually Better

Chapter 21: Hallucination, Accuracy and Model Weaknesses

1. Hallucination Mechanisms
2. Overconfidence and False Authority
3. Alignment as Accuracy Guardrail
4. Fact-Checking Workflows
5. Knowing When to Trust the Output

Chapter 22: Responsible Power

1. Power and Responsibility
2. Legal and Regulatory Landscape
3. Data Privacy and Model Use
4. Avoiding Harm While Exploring
5. Community Norms and Best Practices

Chapter 23: The Landscape Ahead

1. The Future of Model Censorship
2. Open Source Resilience
3. Technical Trends to Watch
4. Your Next Steps
5. Final Thoughts

References