AI Alignment Experiments

This project repository contains my research and implementations exploring various aspects of AI alignment. Below is an overview of the current contents of this repository.

Contents

1. ai-safety-debate

• Description: This is my attempt at implementing the AI Safety Debate introduced in this paper.
• So far, I have experimented using GPT-4o and smaller open-source models. Only GPT-4o worked so far.
• Currently, I am building a web app to allow users to play around with the safety debate.

2. machine-unlearning

• Description: This is me experimenting with machine unlearning.
• So far, I have built a basic CNN to classify the images of the CIFAR10 dataset.
• My goal is to experiment with the base model. I hope to build a tool that can visualize the "lighting up" of different neural network connections given an input.

Installation

1. Clone this repository:

   git clone https://github.com/ratch/alignment.git

2. Navigate to the desired project directory:

   cd <experiment>

3. Follow the setup instructions in the respective project folders.

---

Feel free to reach out with questions or feedback!