paper-buf

Top-10 paper of the day to read

• LLM safety related
  • What is in your data? Identifying benign data that breaks safety
  • Language models learn to mislead humans via rlhf
  • Salora: Safety-alignment preserved low-rank adaptation
  • Level -level self-exposure and patch: Affirmative token mitigation for jailbreak attach defense
  • Turning logic against itself: Probing model defenses through contrastive questions
• Others
  • Predicting the performance of black-box llms through self-queries
  • Dynamical skill adaptation for large language models
  • Grokking at the edge of numerical stability
  • A statistical theory of contrastive pre-training and multimodal generative ai