Preference learning

Sat, 18 Jul 2026 10:11:14 +0000

Definition

Preference learning focuses on teaching models to distinguish between good and bad outputs based on human judgments rather than absolute labels. It typically involves collecting pairs of responses where humans indicate their preferred option. Algorithms then optimize the model to maximize the likelihood of generating preferred responses. This is crucial for aligning large language models with human values, improving safety, and enhancing relevance in conversational AI systems.

Summary

A technique that trains models to align outputs with human preferences using comparative feedback.

Human Feedback on English AI Terms Dictionary

Preference learning

Definition

Summary