RLHF (reinforcement learning from human feedback)
Training a model to produce responses people prefer, using human rankings or ratings of model outputs as the learning signal.
RLHF relies on preference data. Expert preference data in specialist domains is especially scarce.
Related terms
Tell us what you need
Book a free 30-minute consultation. Tell us what data you want to buy or license, and we'll tell you honestly whether we can help and what it involves.
- We reply to every enquiry within two working days.
- No upfront fees for sellers
- We never hold your raw data