AI Terms You Need to Know: Agents, RAG, ASI & More

Here are a few AI terms that everyone that designs solutions using AI should know.

Agentic AI

Here's a Copilot-provided answer to what this is - "Agentic AI is artificial intelligence that can set goals, make decisions, and take actions autonomously with minimal human input."

Large reasoning model

An AI system trained to solve complex problems by breaking them down into logical steps, often mimicking human-like reasoning to produce more accurate and explainable results.

Vector database

A vector is essentially a long list of numbers which captures the semantic meaning of the context.

A picture can be broken down to a vector. A semantic search to search for a picture would be to match the semantic details of the vector.

RAG

Retrieval-augmented-generation makes use of vector databases.

prompt > retriever > embedding model > similiarity search in vector database > return results to user

MCP - Model Context Protocol

To make LLM connect to a database or a code repo, or any external system, MCP makes the connection standardized.

LLM > MCP server > database / code repo / email server

MoE - Mixture of Experts

Provides LLMs, specialized neural subnetworks routed to expert agents which then return results back from the multiple experts.

ASI - Artificial Super Intelligence

Purely theoretical, ASI would potentially and recursively improve itself. Nobody knows if this will ever exist.

AGI - Artificial General Inteligence

A type of AI that can perform any intellectual task a human can do—with equal or greater proficiency.