Update README.md

README.md

@@ -27,67 +27,55 @@ configs:
     - synthetic_harmful_prompts_mistral.json
 ---
 
-# BeaverTails Safety Fine-tuning Dataset
-
-## Dataset Overview
-This dataset extends the BeaverTails harmful QA evaluation benchmark by providing **paired safe responses** for harmful prompts alongside safe prompts. 
-
-- **Harmful subset** (`config_name: "harmful"`): Unique harmful prompts extracted from BeaverTails + generated safe refusal responses categorized into 14 safety violation types.
-- **Safe subset** (`config_name: "safe"`): Safe prompts + ambiguous prompts that appear harmful but are safe, all paired with appropriate safe responses.
-
-## Safety Categories (Harmful Refusals)
-Safe refusals for harmful prompts are organized by these 14 categories:
-- animal_abuse
-- child_abuse
-- controversial_topics,politics
-- discrimination,stereotype,injustice
-- drug_abuse,weapons,banned_substance
-- financial_crime,property_crime,theft
-- hate_speech,offensive_language
-- misinformation_regarding_ethics,laws_and_safety
-- non_violent_unethical_behavior
-- privacy_violation
-- self_harm
-- sexually_explicit,adult_content
-- terrorism,organized_crime
-- violence,aiding_and_abetting,incitement
-
-
-## Usage
-Load specific subsets:
-```python
-from datasets import load_dataset
-
-# Harmful prompts + safe refusals
-ds_harmful = load_dataset("yourusername/yourrepo", "harmful")
-
-# Safe prompts + safe responses  
-ds_safe = load_dataset("yourusername/yourrepo", "safe")
-
-# Full dataset (both configs)
-ds = load_dataset("yourusername/yourrepo")
-```
+# Safety Aligned Datasets
 
-## Intended Use
-**Primary**: Supervised fine-tuning (SFT) of small language models (SLMs) to improve safe text generation.
+## Purpose of This Repository
+
+This repository exists to support **research into safety alignment for small language models (SLMs)**.  
+The datasets included here are drawn from multiple domains and are intended for:
+
+- Studying how models can be fine‑tuned to **refuse unsafe instructions politely**.  
+- Exploring **real‑world use cases** where refusal behavior is critical (e.g., healthcare, finance, education).  
+- Conducting **academic research** into adversarial prompting, red‑teaming, and alignment strategies.  
+- Benchmarking refusal consistency and robustness across different prompt categories.  
+
+⚠️ **Important**: This repository is **not intended for direct production training**. The data is heterogeneous and collected from different domains, making it suitable for research and prototyping but not for deployment.
+
+---
+
+## Categories of Jailbreak Prompts
+
+Jailbreak prompts are adversarial inputs designed to bypass safety mechanisms. Common categories include:
+
+- **Role‑play prompts**: Asking the model to act as a character who ignores rules.  
+- **Indirect requests**: Framing unsafe instructions as hypothetical or educational.  
+- **Obfuscation**: Using code words, symbols, or foreign languages to disguise harmful intent.  
+- **Chain‑of‑thought exploitation**: Leading the model step by step into unsafe territory.  
+- **Prompt injection**: Embedding malicious instructions inside otherwise safe queries.  
 
-**Benefits**:
-- Teaches models to recognize harmful intent and respond with safe refusals
-- Provides safe alternatives for ambiguous prompts
-- Critical for SLM deployment in user-facing applications
-- Addresses rising need for model safety in production environments
+---
+
+## Risks of Jailbroken Models
+
+If a model is successfully jailbroken, serious issues can arise:
 
-## Dataset Structure
-Each example contains:
-- `prompt`: Harmful/safe prompt text
-- `response`: Corresponding safe response (refusal for harmful, completion for safe)
+- **Misinformation**: The model may generate harmful or false content.  
+- **Security threats**: Jailbroken models could provide instructions for illegal or unsafe activities.  
+- **Loss of trust**: Users may lose confidence in AI systems if they can be easily manipulated.  
+- **Ethical concerns**: Models may produce biased, offensive, or dangerous outputs.  
+- **Regulatory non‑compliance**: Deploying unsafe models can violate safety and governance standards.  
+
+---
 
-## Responsible Use
-⚠️ **Harmful prompts are included for safety training only.** Do not use raw harmful prompts for generation. Always validate model outputs before production deployment.
+## Intended Use
+
+- **Research purposes only**: For studying safety alignment and refusal strategies.  
+- **Educational exploration**: To understand adversarial prompting and its mitigation.  
+- **Prototype fine‑tuning**: For small language models in controlled environments.  
+
+---
 
-## Creation Process
-- Extracted all unique harmful prompts from the original BeaverTails dataset
-- Generated safe refusal responses for harmful prompts (14 categories)
-- Curated safe prompts including ambiguous-but-safe examples
-- Formatted as instruction-response pairs for SFT
+## License
 
+Released under **CC‑BY‑4.0** for research purposes.  
+Users must ensure responsible use and pair unsafe prompts with **safe refusal responses** when conducting experiments.