Evals And Observability easy
An LLM judge scores responses from 1 to 10. Scores for the old and new prompt both cluster around 7, and reruns shift by about a point. What change gives a more reliable comparison?
- A Expand the scale to 1 to 100 so the judge can express finer distinctions between similar-quality responses.
- B Show the judge both responses side by side for each input, ask which is better, and count the win rate.
- C Keep absolute scores but have the candidate model grade its own outputs, since it best understands its intent.
- D Average the absolute scores across three runs, then ship if the new prompt's mean is higher by any amount.
Answer: B. Show the judge both responses side by side for each input, ask which is better, and count the win rate.
Pairwise comparison is more reliable than absolute scoring because judges are more consistent at picking the better of two answers than at placing one on a scale; the tradeoff is that you get a relative ranking, not an absolute quality level. A wider 1 to 100 scale is tempting but adds false precision and more noise. Absolute scores still fit when you need a pass/fail threshold on one system rather than a comparison.
One system design question a day, with spaced repetition so the ones you miss come back.
Get the appRelated questions
- In one release, a RAG assistant got a new embedding model and a new chunking strategy, and the end-to-end judge score dropped 6 points. The team only tracks the end-to-end score and has spent a week tweaking the generation prompt with no recovery. What is the best next step?
- Over two months, your team iterated through 40 prompt versions, each tuned against the same 150-case golden set. The offline score rose from 72 to 91. Online task completion and retry rate have not changed. What is the most likely problem, and what is the fix?
- A prompt change produced an impressive answer in a live demo for leadership. The PM wants to ship it today. What should you do?
- You plan to replace human graders with an LLM judge for your answer-quality eval. Before trusting its scores in a CI gate, what should you do first?
- After you swap to a smaller model, p95 latency falls from 900 ms to 350 ms and the ops dashboard is all green. Leadership asks whether answer quality is still fine. What is the right response?
- Your CI gate fails any prompt change that drops the score 3 or more points on a 100-point rubric over 60 cases. Production generates at temperature 0.7. Running the eval twice on the same commit gives scores that differ by up to 4 points, and engineers have started rerunning failed gates until they pass. How do you fix the gate?