LLM-as-a-Judge
Evaluation & MetricsChapter 10: Report Card for BitTerm 173 of 253
Using an LLM to grade another LLM's outputs against criteria.
At Sunrise School
Gopal asksWho grades 200 written replies every night?
Bit answersLLM-as-a-judge: another model scores each reply for accuracy, tone and completeness, from 1 to 5.
In the real worldClaude scores chatbot answers 1–5 for accuracy.
Watch the short
Short coming soonSeason 2 · Day 29Follow @haigopala
Short coming soonSeason 2 · Day 29Follow @haigopala.telugu
Short coming soonSeason 2 · Day 29Follow @haigopala.hindi
Quick check
Quick check. Should humans still spot-check the judge? Yes or no?
Yes! Judges can be wrong too.
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