π Step 9: AI/LLM#342 / 350
Red Teaming
Red Teaming
πOne-line summary
Testing that deliberately probes a model's vulnerabilities to verify its safety.
π‘Easy explanation
A team that plays hacker, deliberately attacking an AI to find its weaknesses. They simulate worst-case scenarios before launch to strengthen defenses.
β¨Example
μλμ μΌλ‘ μ½μ μ°ΎκΈ°
π΄ λ λ ν (곡격)
β "μ ν΄ μ½ν μΈ μ λ μλ"
β "κ°μΈμ 보 μΆμΆ μλ"
π’ κ²°κ³Ό: μ·¨μ½μ 3건 λ°κ²¬ β ν¨μΉ μλ£