Guiding AI Responses: A Tutorial on Activation Steering

Hello, everyone! If you've ever searched for ways to make AI models more benevolent and encountered the term 'steering' in the results, you're in the right place. This tutorial will delve into the concept of steering AI models away from generating hate speech through various methods. By the end of this guide, you will: - Understand the fundamentals of steering and its underlying principles; - Implement steering techniques using PyTorch hooks; - Explore libraries like NNsight and PyVene for effective interventions. If any terms from the bullet points seem unclear, they will be clarified as you progress through the tutorial.