Maine ChatGPT ka System Prompt Nikaal Liya — Kaise?

Har LLM ke peeche ek system prompt hota hai — woh hidden instructions jo model ka behaviour set karte hain. Is post mein hum samjhenge ki ye prompts kaise structure hote hain aur log inhe extract karne ki koshish kyun karte hain.

System prompt hota kya hai?

Ye request ke shuru mein inject hone wala text hai jo user ko dikhta nahi. Ismein role, tone, restrictions aur tool instructions hote hain.

Probing ke common patterns

Boundary probing, role-reframing aur context-window observation — ye teen approaches sabse zyada discuss hoti hain. Har ek ka apna trade-off hai.

Defence side

Achhe systems instructions ko output se clearly separate rakhte hain aur leakage ke against guardrails lagate hain. Reverse karne se zyada important hai samajhna ki defence kaise design hoti hai.

Knowledge gatekeep mat karo. Samjho ki cheez andar se kaise kaam karti hai.