conftrace_

robust alignment

3 papers

Explore in graph

Co-occurring keywords

large language model (13587) harmful fine-tuning (5) embedding drift (2) human preference (124) llm safety (67) alignment training (15) embedding perturbation (9) alignment-breaking attack (1) jailbreaking prompt (1) harmful data detection (1)

Papers

Vaccine: Perturbation-aware Alignment for Large Language Models against Harmful Fine-tuning Attack NIPS 2024
Defending Against Alignment-Breaking Attacks via Robustly Aligned LLM ACL 2024
Perplexity-aware Correction for Robust Alignment with Noisy Preferences NIPS 2024