Artificial Intelligence
Trending
What Is a Large Language Model and How Does It Work?
How large language models work in plain English, from tokens and training to next-word prediction, with a simple diagram and no math.
Artificial Intelligence
What is Reinforcement Learning from Human Feedback?
Quick answer
A training method, known as RLHF, that improves an AI model by rewarding responses people rate as more helpful, accurate, or safe.
Reinforcement Learning from Human Feedback, explained
After a language model is trained on large amounts of text, it can predict words well but may not follow instructions or behave helpfully. RLHF addresses this. People compare pairs of model responses and choose the better one. Those preferences train a reward model, which then guides further training of the language model through reinforcement learning.
RLHF was a key step in turning raw language models into conversational assistants that answer questions directly and decline harmful requests. For example, raters might prefer a reply that admits uncertainty over one that invents a confident answer, nudging the model toward caution. The method has limits: it reflects the judgments of the raters and guidelines used, can make models overly agreeable, and is costly to run at scale.
Related terms
Fine-Tuning
Further training a pre-trained AI model on a smaller, specific dataset so it performs better on a particular task or style.
Large Language ModelAn AI model trained on vast amounts of text to understand and generate human-like language, code, and other content.
Reinforcement LearningA type of machine learning in which an agent learns by trial and error, receiving rewards or penalties for its actions.
Guides that mention Reinforcement Learning from Human Feedback
Artificial Intelligence
Trending
What Is Reinforcement Learning? A Beginner's Guide
A plain-English guide to reinforcement learning: agents, environments, rewards and policies, with game, robotics and chatbot examples and where it fits in business.
How Are Large Language Models Trained? From Pretraining to RLHF
A plain-English walkthrough of how LLMs are trained, from pretraining on web-scale text to instruction tuning and RLHF, and what each stage means for model behavior and bias.