AI Glossary: Key Definitions

RLHF is a machine learning technique used to fine-tune AI models using human feedback to improve their behavior and outputs. The key steps involved are:

• Response Generation: The AI generates a set of responses to a given input.
• Human Feedback: Humans rank or provide feedback on the quality of the responses.
• Training: The AI is trained using reinforcement learning to favor responses that align with human preferences.

Applications: RLHF is used to align AI with human values, reduce harmful outputs, and ensure responses are more relevant, ethical, and understandable.

RAG is an AI architecture that combines a language model with information retrieval to provide factually grounded responses. The process involves:

• Information Retrieval: The AI pulls relevant data from external knowledge bases.
• Augmentation: The retrieved information is fed into the language model to improve accuracy and relevance.
• Response Generation: The AI generates a final output using both retrieved information and its existing knowledge.

Applications: RAG is commonly used in chatbots, virtual assistants, and question-answering systems to improve the factual accuracy and relevance of responses.

The process of adapting a pre-trained model to a specific task using additional training data. It helps specialize a general AI model for niche tasks or industries.

Designing inputs or queries to guide AI models toward producing desired outputs. Effective prompt engineering can improve response quality without changing the model itself.

A broader machine learning approach where agents learn by interacting with an environment and receiving rewards or penalties. This feedback loop encourages the agent to take actions that maximize rewards over time.

A structured repository of information used by AI systems for information retrieval. In RAG systems, the knowledge base is queried to provide factually grounded outputs.

A method where humans remain involved