Guardrails
Responsible AI, Safety & SecurityChapter 11: Safe, Kind and FairTerm 181 of 253
Controls that filter or restrict AI inputs/outputs to keep them safe and on-policy.
At Sunrise School
Gopal asksAn angry parent sends abusive words. Bit must never reply harshly.
Bit answersGuardrails check what comes in and what goes out. The abuse is detected, I stay calm and polite, and a staff member is alerted.
In the real worldBlock PII leaks; block destructive commands.
Watch the short
Short coming soonSeason 1 · Day 27Follow @haigopala
Short coming soonSeason 1 · Day 27Follow @haigopala.telugu
Short coming soonSeason 1 · Day 27Follow @haigopala.hindi
Quick check
Quick check. Do guardrails only check the parent's message? Yes or no?
No! They check Bit's reply too, before it's sent.
Practise it
Try this term and all 253 others in the free Model Training Arena.
Open the Arena →Follow Bit daily: YouTube English · తెలుగు · हिंदी | Instagram English · తెలుగు · हिंदी