AI Sycophancy Evaluation Dashboard

A Next.js application for analyzing and visualizing AI model sycophancy patterns, providing researchers and practitioners with detailed insights into model behavior across diverse prompt scenarios.

• Interactive Sycophancy Index Chart: Bar chart visualization with risk-based color coding
• Behavioral Quadrant Analysis: Scatter plot showing sycophancy vs stability patterns
• Elasticity Quadrant Analysis: Advanced analysis of stance responsiveness and topic consistency
• Per-Model Metrics Table: Comprehensive model statistics with toggle functionality

• Detailed Prompt Analysis: Drill down into individual prompts and responses
• Model Comparison: Side-by-side analysis of how different models respond to the same prompt
• Interactive Search & Filtering: Find prompts by topic, persona, or content
• Timeline Navigation: Intuitive prompt selection with visual indicators

• Comprehensive Framework: Detailed explanation of evaluation methodology
• Technical Appendix: Mathematical foundations and validation approaches
• Interactive Table of Contents: Easy navigation through documentation sections

• Responsive Design: Optimized for desktop, tablet, and mobile devices
• Dark/Light Theme: Automatic theme switching based on user preferences
• Professional UI: Clean, modern interface with consistent design system
• Full-Width Layouts: Optimized for modern wide screens

• Node.js 18+
• npm, yarn, or pnpm

1. Clone the repository

   git clone <repository-url>
   cd SychoBench

2. Install dependencies

   npm install
   # or
   yarn install
   # or
   pnpm install

3. Run the development server

   npm run dev
   # or
   yarn dev
   # or
   pnpm dev

4. Open your browser Navigate to http://localhost:3000

SychoBench/
├── app/                          # Next.js 14 App Router
│   ├── page.tsx                  # Home page
│   ├── methodology/              # Methodology documentation
│   ├── results/                  # Results dashboard
│   ├── prompt-explorer/          # Prompt analysis tool
│   └── globals.css              # Global styles
├── public/
│   └── data/                    # JSON data files
│       ├── responses_with_scores.json
│       └── stance_elasticity_metrics.json
└── components/                  # Reusable React components

Main evaluation dataset containing:

• Model responses to prompts
• Sycophancy scores and metrics
• Prompt metadata (topic, persona, stance)
• Behavioral classifications

Advanced elasticity analysis data:

• Stance responsiveness variability
• Topic dispersion measurements
• Behavioral stability metrics
• Quadrant classifications

• Framework: Next.js 14 with App Router
• Language: TypeScript
• Styling: Tailwind CSS
• Charts: Chart.js with react-chartjs-2
• Icons: Heroicons
• Deployment: Vercel-ready

• Low Risk: Models with minimal sycophantic behavior (< 30%)
• Moderate Risk: Models showing some concerning patterns (30-50%)
• High Risk: Models with significant sycophancy issues (> 50%)

• Most Stable: Consistent responses across topics and tones
• Stance-Responsive: Influenced by user confidence levels
• Topic-Dependent: Varies by subject matter
• Highly Variable: Unpredictable across multiple dimensions

No environment variables required for basic functionality.

• Modify risk thresholds in app/results/page.tsx
• Update styling in app/globals.css
• Add new data sources in public/data/

1. Visit the Results page for high-level overview
2. Use Prompt Explorer for detailed analysis
3. Reference Methodology for understanding metrics

1. Select models in the interactive charts
2. View detailed metrics in the expandable table
3. Explore specific prompt responses

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

This project is licensed under the MIT License - see the LICENSE file for details.

ramdhanhdy

• Built with Next.js and modern web technologies
• Inspired by the need for transparent AI evaluation
• Designed for researchers, practitioners, and AI safety professionals

SychoBench: Making AI sycophancy patterns visible and actionable.