A list of existing alignment approaches

LessWrong has published a comprehensive taxonomy of AI alignment approaches—the techniques researchers are using to train AI systems toward helpfulness and safety. The post catalogs training methods ranging from internals-based approaches that use model reasoning as a reward signal, to output-focused techniques like reinforcement learning. It covers everything from imitation-based training to teaching AI systems through stories about alignment, plus meta-approaches like ensemble systems that combine multiple AI models and debate-style techniques for interrogating whether an AI is trying to sabotage you. The central challenge, as the post frames it, is turning this menu of techniques into a real, trustworthy AI system.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton