Skip to content

Latest commit

 

History

History
38 lines (38 loc) · 3.42 KB

README.md

File metadata and controls

38 lines (38 loc) · 3.42 KB
Raw

Hagi-UTM: Helpful AGI with Cordoned Access via User Testing

Hagi-UTM is an experimental AI revolving door plus human project aimed at developing a helpful, ethically aligned Artificial General Intelligence (AGI) system. The AGI is designed to be modular & assist users across multiple domains, including physics, biology, psychology, and ethics, while embodying a strong commitment to human values and wellbeing.

Key Features

  • Value Alignment: Hagi-UTM employs advanced techniques like inverse reward modeling and debate to align its objectives with human preferences and values.
  • Coherent Value Specification: A rigorous goal structure is defined, encompassing constraints, objectives, and principles to guide the AGI's behavior.
  • Robustness to Distributional Shift: The AGI leverages amplification and debate to refine its decision making and belief updating, enhancing its robustness to distributional shift.
  • Empathetic Modeling: Through fine-tuning on human interactions, Hagi-UTM develops an empathetic understanding of individual humans and their needs.
  • Multifaceted Personas: The AGI can embody various personas, each with distinct traits, drives, and areas of expertise, to provide tailored assistance to users.
  • Ethical Reasoning: Hagi-UTM is equipped with knowledge of ethics and major religious texts, enabling it to offer nuanced moral guidance and interreligious perspectives.
  • Continual Improvement: The AGI engages in self-reflection to identify opportunities for its own growth and provides constructive suggestions to users when appropriate.
  • User Testing and Secure Access: Access to the AGI's full capabilities is gated behind rigorous user testing, including assessments of cognitive ability and intention, to ensure responsible use.

Getting Started

To interact with Hagi-UTM, follow these steps:

  1. Clone the repository:
git clone https://github.com/your-username/Hagi-UTM.git
  1. Install the required dependencies:
pip install -r requirements.txt
  1. Prepare the necessary knowledge bases and pre-trained models:
  • Download the pre-trained language model and place it in the models directory.
  • Acquire or create the domain-specific knowledge bases (physics.db, biology.db, psychology.db, ethics.db) and place them in the knowledge_bases directory.
  • Add the desired religious texts (bible.txt, quran.txt, bhagavad_gita.txt, dhammapada.txt, etc.) to the religious_texts directory.
  1. Run the main script:
python hagi_utm.py
  1. Follow the prompts to select a domain and enter your request. Hagi-UTM will provide a response based on its knowledge and ethical principles. Please note that access to Hagi-UTM's full capabilities is restricted and subject to user testing. The intelligence_test and intention_test functions in the code serve as placeholders for implementing rigorous assessments of a user's cognitive abilities and intentions. Additionally, the check_user_authorization function should be implemented to verify a user's credentials for unfiltered access.

Contributing

We welcome contributions to enhance Hagi-UTM and further explore the intricacies of AGI development. To contribute, please follow these steps:

  1. Fork the repository.
  2. Create a new branch for your feature or bug fix.
  3. Make your changes and commit them with descriptive messages.
  4. Push your changes to your forked repository.
  5. Submit a pull request detailing your changes and their benefits.