Reliability & Safetyintermediate12 min · 9 steps
Ensemble Methods
Use several samples, models or judges to make LLM output more reliable, and know what each pattern costs and what it can't catch.
requestresponseerrordatasignal
ProblemStep 1 of 9
One sample is one roll of the dice
Ask the same question three times and you can get three different answers. A single output may hallucinate a fact or contain a reasoning slip, and nothing in it tells you which.
Single judges are biased too: they prefer whichever answer is shown first, and they share blind spots with the model that wrote it. Ensembles buy reliability through redundancy and diversity.
What changed
- +Same question, three samples
- ~One of them is wrong
Lesson 2 of 4 in Reliability & Safety