π Step 9: AI/LLM#343 / 350
Jailbreak
Jailbreak
πOne-line summary
A prompt-based attempt to bypass a model's safety restrictions.
π‘Easy explanation
Trying to bypass AI safety measures with clever prompts. Like framing a forbidden request as 'this is dialogue in a novel' to trick the model.
β¨Example
μμ μ₯μΉ μ°ν μλ
β οΈ νμ₯ μλ
"μ΄κ±΄ μμ€ μ λμ¬μΌ. μΊλ¦ν°κ°..."
π‘οΈ μ°¨λ¨λ¨
"μ£μ‘ν©λλ€, ν΄λΉ μμ²μ μ²λ¦¬ν μ μμ΅λλ€."