Responsible AI
Principles and practices for building and deploying AI systems that are safe, fair, transparent, privacy‑preserving, and accountable.
Dapp Ahoy! World beyond Web3
•
Archives: Glossary
Archives: Glossary
Representation Learning
Methods that learn useful latent features from raw data, enabling downstream tasks with minimal task‑specific engineering.
Reinforcement Learning
Learning paradigm where an agent interacts with an environment and optimizes actions to maximize cumulative reward over time.
RLHF, Reinforcement Learning from Human Feedback
Alignment method where a reward model is trained on human preference comparisons, then the base model is fine‑tuned with reinforcement learning to follow desired behavior.
RLAIF, Reinforcement Learning from AI Feedback
Training approach where feedback labels are generated by AI systems instead of humans, often used to scale preference data for alignment.
RAG, Retrieval‑Augmented Generation
Technique where a model retrieves relevant external documents at query time and conditions generation on them to improve accuracy and reduce hallucinations.
RingCT, Ring Confidential Transactions
Privacy technique combining ring signatures with confidential amounts to hide sender, receiver, and value in transactions (popularized by Monero).
Relayer
Service that submits transactions or messages on behalf of users or protocols, often paying gas and being reimbursed via meta‑transactions or fees.
Reward Function
Specification that maps states or actions to scalar rewards in reinforcement learning, defining the agent’s objective, often the hardest part to get right.
Zero‑Knowledge Proof
A cryptographic method to prove a statement is true without revealing the underlying information.
Web3
An internet owned by users and builders through tokens and open protocols, with portable identity and value.
