RLHF
Technology & Computing abbreviation · Artificial Intelligence
| Stands for | Reinforcement learning from human feedback |
|---|---|
| Categories | Technology & Computing |
RLHF stands for Reinforcement learning from human feedback, an artificial intelligence abbreviation.
A training method in which people compare or rate a model's outputs, a reward model learns those preferences, and the language model is then optimized with reinforcement learning to produce responses people prefer. It made chatbots far more helpful and better at following instructions.
Example: After RLHF, the model stopped rambling and started answering the actual question.
Usage notes
Early work came from OpenAI and DeepMind researchers in 2017 on learning from human preferences, and OpenAI's InstructGPT paper (2022) applied it to language models shortly before ChatGPT. Human labeling is expensive and can encode rater biases. Alternatives include RLAIF, which uses AI feedback, and direct preference optimization, which skips the separate reward model.
Related abbreviations
More artificial intelligence abbreviations
- QA – Question answering
- RAG – Retrieval-augmented generation
- ReLU – Rectified linear unit
- RF – Random forest
- RL – Reinforcement learning
- RLAIF – Reinforcement learning from AI feedback
- RMSE – Root mean square error
- RNN – Recurrent neural network
- ROC – Receiver operating characteristic
- RPA – Robotic process automation
- SD – Stable Diffusion
- SFT – Supervised fine-tuning