Alignment & Safety

Methods to align models with human preferences and constraints.

Topics

• RLHF/DPO and preference data
• Guardrails, policy engines, allow/deny lists
• Red teaming and jailbreak resilience
• Privacy, security, and compliance