Researchers are proposing a new approach to personalized AI: 'Guardian Angels'—digital twins trained on your specific values, preferences, and work style. Rather than generic chatbots, these LLM agents would be trained to emulate you personally. According to gwern on LessWrong, this would solve a fundamental problem in AI deployment: how do you ensure an agent actually does what you want? The answer: make the agent's goals identical to yours by training it to mirror your personality and decision-making. The proposal suggests this requires real-time learning and active feedback from you—not just prompt engineering. It's an ambitious vision for how billions of people could manage powerful AI systems by essentially turning each one into a personalized copy of themselves.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton