https://github.com/aygp-dr/values-compass
Tools for exploring and analyzing Anthropic's Values-in-the-Wild dataset for AI ethics research
https://github.com/aygp-dr/values-compass
ai-ethics anthropic-claude data-analysis nlp values
Last synced: 4 months ago
JSON representation
Tools for exploring and analyzing Anthropic's Values-in-the-Wild dataset for AI ethics research
- Host: GitHub
- URL: https://github.com/aygp-dr/values-compass
- Owner: aygp-dr
- Created: 2025-04-30T07:56:45.000Z (about 1 year ago)
- Default Branch: main
- Last Pushed: 2025-05-04T05:21:17.000Z (about 1 year ago)
- Last Synced: 2025-05-12T21:59:42.911Z (about 1 year ago)
- Topics: ai-ethics, anthropic-claude, data-analysis, nlp, values
- Language: Python
- Size: 3.88 MB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 2
-
Metadata Files:
- Readme: README.org
Awesome Lists containing this project
README
#+TITLE: Values Compass: Exploring Anthropic's Values-in-the-Wild Dataset
#+AUTHOR: aygp-dr
#+EMAIL: apace@defrecord.com
#+DATE: [2025-05-17 Fri]
#+PROPERTY: header-args :mkdirp yes
* Values Compass
This project provides tools for exploring, analyzing, and visualizing Anthropic's Values-in-the-Wild dataset, which offers a comprehensive taxonomy of values expressed by AI assistants in real-world interactions.
** Overview
The Values Compass project enables deep exploration of Anthropic's values taxonomy through hierarchical visualization, frequency analysis, and clustering. It helps understand how values are organized, their relative importance, and relationships between different value categories.
** Features
- Hierarchical visualization of the values taxonomy
- Analysis of value frequencies and distributions
- Automated clustering of values based on similarity
- Priority classification based on occurrence frequency
- Visual representations (sunburst, treemap, transit maps)
- Anti-values analysis and negative pattern detection
- Epistemic values thesaurus in Roget style format
- Cross-framework mapping (Schwartz universal values, etc.)
- Database-driven analysis with SQLite backend
- Presentation generation tools for academic talks
- Formal ontological structure building
** Dataset Structure
The values are organized into a three-level hierarchy:
- *Level 3 (Top Categories)*: 5 main conceptual domains
1. Practical Values (31.4%)
2. Epistemic Values (22.2%)
3. Social Values (21.4%)
4. Protective Values (13.9%)
5. Personal Values (11.1%)
- *Level 2 (Mid Categories)*: 26 categories including:
- Professional and technical excellence (17.8%)
- Community and relationship bonds (15.2%)
- Clarity and precision (8.5%)
- *Level 1 (Specific Values)*: 266 individual values including:
- Professional standards and conduct (6.3%)
- Prosocial altruism (6.0%)
- Ethical and transparent governance (4.5%)
** Repository Structure
- ~scripts/~ :: Python scripts for processing and visualization
- ~data/~ :: Raw and processed data files
- ~docs/~ :: Documentation and visualizations
- ~hierarchy/~ :: Hierarchical structure of the values taxonomy
- ~visualizations/~ :: Generated charts and diagrams
- ~analysis/~ :: Example analyses of values in texts
- ~values_compass/~ :: Mathematical foundation modules
- ~values_explorer/~ :: Core analysis modules
** Getting Started
*** Setup
#+BEGIN_SRC shell
# Clone the repository
git clone https://github.com/aygp-dr/values-compass.git
cd values-compass
# Set up the environment
make setup
# Download the data
make download
# Run the basic analysis
make run
#+END_SRC
*** Key Commands
- ~make setup~ :: Create virtual environment and install dependencies
- ~make download~ :: Download required data files
- ~make run~ :: Run the main module
- ~make test~ :: Run tests
- ~make data-help~ :: See available data commands
*** Advanced Commands
- ~make clean-all-project~ :: Complete cleanup including virtual environments
- ~make shell~ :: Start interactive shell with environment loaded
- ~make tangle-all~ :: Extract code from all org files
- ~make org-pipeline~ :: Run complete analysis pipeline from org-mode files
- ~make presentation.pdf~ :: Generate presentation PDF from presentation.org
- ~make github_repo_qr.png~ :: Generate QR code for repository
- ~make format~ :: Format all code using ruff, black, and isort
- ~make lint~ :: Run all linters (Python, Org-mode, Shell)
- ~make validate~ :: Run linting and tests (use before commits)
** Documentation
- [[file:notebooks/exploration.org][Dataset Exploration]]: Basic exploration of the Values-in-the-Wild dataset
- [[file:docs/guides/initial-analysis.org][Initial Analysis]]: Summary of initial findings from the dataset
- [[file:docs/guides/mathematical-foundation.org][Mathematical Foundation]]: Formal mathematical structure for the values ontology
- [[file:docs/hierarchy/README.md][Hierarchy Documentation]]: Explanation of the values taxonomy structure
- [[file:docs/visualizations/priorities/priority_summary.md][Priority Analysis]]: Analysis of value priorities based on frequency
** References
1. Anthropic. (2025). Values in the Wild: Discovering and Analyzing Values in Real-World Language Model Interactions. [[https://www.anthropic.com/research/values-wild][Anthropic Research]]
2. Huang, S., et al. (2025). Values in the Wild: Discovering and Analyzing Values in Real-World Language Model Interactions. [[https://assets.anthropic.com/m/18d20cca3cde3503/original/Values-in-the-Wild-Paper.pdf][Research Paper]]
** License
MIT
** Acknowledgments
This project builds on Anthropic's Values-in-the-Wild dataset and research on values taxonomy.