R

A class of AI models specifically trained to think through problems step-by-step before producing an answer, dramatically improving performance on complex reasoning, math, and coding tasks. OpenAI's o1 and o3, DeepSeek R1, and Gemini 2.0 Flash Thinking are reasoning models. They trade response speed for accuracy on hard problems.

A machine learning paradigm where an AI agent learns by taking actions in an environment and receiving rewards or penalties based on the outcomes. The agent learns to maximize cumulative reward over time. Reinforcement learning is the foundation of game-playing AI (AlphaGo, AlphaZero) and is used in RLHF to align LLMs with human preferences.

A training technique used to align AI models with human preferences. Human raters evaluate AI outputs and their preferences are used to further train the model to produce more helpful, harmless, and honest responses. RLHF is a key technique used by OpenAI, Anthropic, and Google to make their models safer and more useful.

Replit's AI coding assistant integrated into its cloud-based development environment. Replit AI can generate code, explain errors, and build complete applications from natural language descriptions. It is particularly popular for beginners and for rapid prototyping of AI-powered applications.

A framework for developing and deploying AI systems in ways that are ethical, transparent, fair, and accountable. Responsible AI principles include fairness (avoiding discriminatory outcomes), explainability (being able to explain AI decisions), privacy (protecting user data), and safety (preventing harm). Major tech companies and governments have published responsible AI guidelines.

A technique that improves AI accuracy by first retrieving relevant information from a knowledge base or document store, then using that retrieved context to generate a more accurate, grounded response. RAG reduces hallucinations and allows AI to answer questions about proprietary or recent information not in its training data.

Software that automates repetitive, rule-based digital tasks by mimicking human interactions with computer interfaces — clicking buttons, copying data, filling forms. RPA is the predecessor to AI automation. When combined with AI (intelligent automation), RPA can handle unstructured data and make decisions, not just follow fixed rules.

A prompting technique where you assign the AI a specific persona or role before asking your question, to shape the style, tone, and expertise of its response. Telling the AI 'You are an expert commercial real estate attorney' before asking a legal question produces more specialized, authoritative responses than asking without context.

A leading AI video generation and editing platform that offers text-to-video, image-to-video, and video editing tools powered by AI. Runway's Gen-3 Alpha model is widely used by filmmakers, marketers, and content creators for professional AI video production.