Conceptpreference alignment
Preference Alignment
An AI alignment technique that adjusts model behavior by collecting human preference feedback on model outputs, such as rankings and ratings, so that outputs better conform to human expectations and domain norms. It is often combined with reinforcement learning (RLHF).
Core Facts
Timeline (last 90 days)
Nov 13
In the context of generative AI, HITL is often combined with preference alignment, that is, using human feedback (such as rankings or ratings in RLHF) to adjust model behavior.
Unverified50%
All Facts (2)
Verified
Large model training goes through three core stages: pre-training, supervised fine-tuning (SFT), and preference alignment.
75%UnverifiedIn the context of generative AI, HITL is often combined with preference alignment, that is, using human feedback (such as rankings or ratings in RLHF) to adjust model behavior.
50%