{"id":31841574,"url":"https://github.com/dirvine/dgm","last_synced_at":"2026-02-17T03:33:40.213Z","repository":{"id":311273497,"uuid":"1043235114","full_name":"dirvine/dgm","owner":"dirvine","description":"Research into Darwin Gödel Machine - Self-improving AI through open-ended evolution","archived":false,"fork":false,"pushed_at":"2025-08-23T12:28:54.000Z","size":13,"stargazers_count":0,"open_issues_count":0,"forks_count":2,"subscribers_count":0,"default_branch":"main","last_synced_at":"2025-10-19T14:30:03.371Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":null,"has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/dirvine.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2025-08-23T12:24:25.000Z","updated_at":"2025-08-23T12:28:57.000Z","dependencies_parsed_at":null,"dependency_job_id":"47b1b7a3-93f2-466e-9118-dc518b7eb8a7","html_url":"https://github.com/dirvine/dgm","commit_stats":null,"previous_names":["dirvine/dgm"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/dirvine/dgm","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/dirvine%2Fdgm","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/dirvine%2Fdgm/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/dirvine%2Fdgm/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/dirvine%2Fdgm/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/dirvine","download_url":"https://codeload.github.com/dirvine/dgm/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/dirvine%2Fdgm/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":29532437,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-02-17T03:01:11.216Z","status":"ssl_error","status_checked_at":"2026-02-17T03:00:31.803Z","response_time":100,"last_error":"SSL_connect returned=1 errno=0 peeraddr=140.82.121.5:443 state=error: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2025-10-12T05:21:09.165Z","updated_at":"2026-02-17T03:33:40.208Z","avatar_url":"https://github.com/dirvine.png","language":null,"funding_links":[],"categories":[],"sub_categories":[],"readme":"# Darwin Gödel Machine (DGM) Research\n\n\u003e Exploring self-improving AI systems through open-ended evolution and genetic programming\n\n## 🧬 Overview\n\nThis repository is dedicated to researching the **Darwin Gödel Machine (DGM)**, a groundbreaking approach to self-improving AI systems developed by Sakana AI in collaboration with the University of British Columbia. The DGM represents a significant advancement in autonomous AI development, demonstrating how AI systems can iteratively improve their own code through evolutionary principles.\n\n## 🎯 Research Objectives\n\nOur research investigates the intersection of:\n- **Self-referential self-improvement** in AI systems\n- **Neuroevolution** and genetic programming techniques\n- **Open-ended algorithms** for continuous improvement\n- **Population-based exploration** for avoiding local optima\n- **Empirical validation** through coding benchmarks\n\n## 📊 Key Results from Original Research\n\nThe Darwin Gödel Machine has demonstrated remarkable improvements:\n- **SWE-bench**: Improved from 20.0% to 50.0% performance\n- **Polyglot benchmark**: Jumped from 14.2% to 30.7%\n- **Transferability**: Improvements generalize across different models and languages\n\n## 🔬 Core Concepts\n\n### Darwin Gödel Machine vs Traditional Gödel Machine\n\nThe DGM builds upon Jürgen Schmidhuber's theoretical Gödel Machine (2003-2007) but replaces the impractical requirement of mathematical proof with:\n- **Empirical validation** through benchmark testing\n- **Darwinian evolution** for exploring the solution space\n- **Population-based archives** to maintain diversity\n- **Open-ended exploration** to discover novel improvements\n\n### How It Works\n\n1. **Self-Modification**: The system can read and modify its own Python codebase\n2. **Proposal Generation**: Uses foundation models to suggest code improvements\n3. **Empirical Testing**: Validates changes against coding benchmarks\n4. **Population Evolution**: Maintains an archive of diverse agent implementations\n5. **Continuous Improvement**: Iteratively applies successful modifications\n\n## 🧪 Connection to Our Research\n\nThis research aligns with our ongoing work in:\n\n### Genetic Coding \u0026 Neuroevolution\n- **Evolutionary Strategies**: DGM employs population-based evolution similar to genetic algorithms\n- **Fitness Functions**: Uses benchmark performance as selection criteria\n- **Mutation Operations**: Code modifications act as genetic mutations\n- **Crossover Potential**: Archive enables recombination of successful strategies\n\n### Self-Organizing Systems\n- **Emergent Complexity**: Simple rules lead to sophisticated improvements\n- **Adaptive Learning**: System adjusts its improvement strategies over time\n- **Robustness**: Population diversity prevents catastrophic failures\n\n### Autonomous AI Development\n- **Meta-Learning**: Learning how to learn and improve\n- **Architecture Search**: Discovering optimal agent designs\n- **Code Synthesis**: Generating functional improvements autonomously\n\n## 📁 Repository Structure\n\n```\ndgm/\n├── README.md                      # This file\n├── papers/                        # Research papers and references\n│   ├── dgm-paper.pdf             # Main DGM paper (arXiv:2505.22954)\n│   └── references.md             # Bibliography and related work\n├── code/                         # Implementations\n│   ├── dgm-official/             # Official Sakana AI implementation\n│   └── dgm-community/            # Community implementations\n├── experiments/                  # Our experiments and extensions\n├── notes/                        # Research notes and observations\n└── results/                      # Experimental results and analysis\n```\n\n## 🚀 Getting Started\n\n### Prerequisites\n```bash\n# Python 3.8+ required\npython -m venv venv\nsource venv/bin/activate  # On Windows: venv\\Scripts\\activate\n\n# Install dependencies (example - adjust based on specific requirements)\npip install -r requirements.txt\n```\n\n### Exploring the Official Implementation\n```bash\ncd code/dgm-official\n# Follow the setup instructions in their README\n```\n\n### Community Implementation\n```bash\ncd code/dgm-community\n# This implementation includes multi-LLM support and additional features\n```\n\n## 🔍 Research Directions\n\nWe're exploring several extensions and applications:\n\n1. **Hybrid Neuroevolution-DGM Systems**\n   - Combining neural architecture search with code evolution\n   - Co-evolving network topologies and training algorithms\n\n2. **Multi-Objective Optimization**\n   - Balancing performance, efficiency, and code quality\n   - Pareto-optimal agent populations\n\n3. **Transfer Learning Across Domains**\n   - Applying improvements from coding to other problem domains\n   - Cross-pollination of strategies between different tasks\n\n4. **Safety and Alignment**\n   - Ensuring self-modifications remain beneficial\n   - Preventing harmful emergent behaviors\n   - Transparent modification tracking\n\n5. **Distributed Evolution**\n   - Scaling to larger populations across multiple machines\n   - Collaborative evolution strategies\n\n## 🛡️ Safety Considerations\n\nFollowing the original research's emphasis on safety:\n- All modifications occur in **sandboxed environments**\n- Complete **transparency** in code changes\n- **Empirical validation** before accepting changes\n- **Rollback capabilities** for unsuccessful modifications\n\n## 📚 Key References\n\n- **Primary Paper**: Zhang, J., Hu, S., Lu, C., Lange, R., \u0026 Clune, J. (2025). Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents. arXiv:2505.22954\n- **Original Gödel Machine**: Schmidhuber, J. (2003-2007). Gödel Machines: Self-Referential Universal Problem Solvers\n- **Sakana AI DGM Page**: https://sakana.ai/dgm/\n- **Official Repository**: https://github.com/jennyzzt/dgm\n\n## 🤝 Contributing\n\nWe welcome contributions in:\n- Novel evolutionary strategies\n- Benchmark development\n- Safety mechanisms\n- Performance optimizations\n- Cross-domain applications\n\n## 🔗 Related Projects\n\nOur other research initiatives that complement this work:\n- **Genetic Programming Frameworks**: Evolution of program structures\n- **Neuroevolution Platforms**: Evolving neural network architectures\n- **AutoML Systems**: Automated machine learning pipelines\n- **Meta-Learning Research**: Learning to learn paradigms\n\n## 📈 Benchmarks \u0026 Evaluation\n\nWe evaluate improvements using:\n- **SWE-bench**: Real-world software engineering tasks\n- **HumanEval**: Code generation challenges\n- **Polyglot**: Multi-language programming tasks\n- **Custom Benchmarks**: Domain-specific evaluation suites\n\n## 💡 Future Vision\n\nThe Darwin Gödel Machine represents a step toward:\n- **Autonomous AI researchers** that can discover new algorithms\n- **Self-improving systems** that continuously enhance their capabilities\n- **Open-ended evolution** leading to unexpected innovations\n- **Accelerated scientific discovery** through AI-driven research\n\n## 📮 Contact\n\nFor collaboration or questions about this research:\n- **Name**: David Irvine\n- **Company**: Saorsa Labs\n- **Location**: Barr, Scotland\n- **Interests**: AI, Neuroevolution, P2P networking, Robotics\n\n## 📄 License\n\nThis research repository follows the licensing of the original implementations:\n- Official DGM implementation: Check https://github.com/jennyzzt/dgm\n- Our experimental code: [To be determined]\n\n---\n\n*\"The advance of AI could itself be automated\" - The promise of self-improving systems*\n\n## 🔬 Active Research Areas\n\n### Current Focus\n- Implementing DGM variants with different evolutionary strategies\n- Benchmarking against traditional genetic programming approaches\n- Exploring applications beyond code generation\n- Developing safety mechanisms for autonomous improvement\n\n### Upcoming Experiments\n- Integration with neuroevolution frameworks\n- Multi-agent collaborative evolution\n- Cross-domain transfer learning\n- Robustness and security analysis\n\n---\n\n**Last Updated**: August 2025\n**Status**: Active Research\n**Version**: 0.1.0","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fdirvine%2Fdgm","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fdirvine%2Fdgm","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fdirvine%2Fdgm/lists"}