AI agent glossary
Ensembles and majority voting
Running the same task several times or several ways and combining the answers, for example by majority vote, so independent mistakes cancel out.
A model asked the same hard question five times will not always give the same answer. Ensembles use that variation. For questions with a definite answer, majority voting picks the most common result. For open-ended work, best-of-n generates several candidates and a judge picks the best. In structured debate, agents critique each other before a decision.
The amount of disagreement is itself useful: a five-to-zero vote deserves more trust than three-to-two, and close votes can be routed to a person.
Ensembles only help when the attempts differ in meaningful ways, such as different framings or different models, and every one multiplies your token bill. Measure whether the accuracy gain is worth the cost.
A classifier runs each contract clause through three differently worded prompts. When all three agree, the label is accepted; when they split, the clause goes to a lawyer.
Learn more in the course
Related terms
Go from definitions to doing
Learn to use and build AI agents in 130 practical lessons.