jailbreaks
2 articles
- ai safetyAI Safety and Character: Can It Be Trained Into a Model?AI safety and character asks whether a model can be safe because of what it is, not only because its outputs are policed, and what would show it.
- behaviorismWhat Is Behaviorism in AI Alignment? Outputs vs CharacterBehaviorism in AI alignment means shaping only what a model outputs, through rewards, refusals and filters, while its interior stays a black box.