Resources - AI Glossary

AGI

Artificial General Intelligence — a hypothetical AI that can perform any intellectual task a human can.

Also known as: General AI

AI

Artificial Intelligence — systems that perform tasks requiring human-like intelligence such as perception, reasoning, and learning.

Alignment

The process of ensuring model behavior aligns with human values and desired outcomes, including safety and ethics.

Context Window

Maximum number of tokens a model can attend to in a single request.

Drift

Changes in data or model behavior over time that degrade performance; includes data and concept drift.

Embedding

A numeric vector representation of text (or other data) used for similarity search, clustering, and retrieval.

Few-shot

Model is given a few in-context examples to better perform a task.

Fine-tuning

Adapting a pretrained model to a narrower task or domain using additional labeled examples.

Guardrails

Controls and validation layers around a model to constrain behavior, format, and safety.

Hallucination

A confident but incorrect output from a model that is not grounded in source data.

Inference

The act of running a trained model to make predictions or generate outputs.

Latency

Time taken from request to response; a key metric in model serving and UX.

LLM

Large Language Model — a transformer-based model trained on large corpora to understand and generate text.

Also known as: Large Language Model

MLOps

Practices and tooling for developing, deploying, monitoring, and maintaining ML systems in production.

Prompt

The input instructions or examples given to a model to elicit a desired response.

Prompt Engineering

Systematically designing prompts (instructions, examples, constraints) to achieve reliable model outputs.

RAG

Retrieval-Augmented Generation — technique that retrieves relevant context (e.g., from a vector DB) and feeds it to a model to improve answers.

Also known as: Retrieval Augmented Generation

Reinforcement Learning

Training agents via rewards to learn sequences of actions that maximize long-term returns.

Retrieval

The process of selecting relevant documents or passages from a corpus to support a model’s answer.

Supervised Learning

Training models using labeled examples mapping inputs to desired outputs.

Temperature

Sampling parameter that controls randomness; higher values produce more diverse outputs.

Token

A subword unit used by language models for processing and counting context length.

Top-k

Sampling technique selecting from the top k most probable tokens.

Top-p

Nucleus sampling that selects the smallest set of tokens whose cumulative probability exceeds p.

Transformer

A neural network architecture using attention mechanisms; foundation of modern LLMs.

Unsupervised Learning

Learning patterns from unlabeled data such as clustering or dimensionality reduction.

Vector Database

A database optimized for storing and searching embedding vectors via similarity metrics.

Also known as: Vector DB

Zero-shot

Model performs a task without seeing annotated examples during training or prompting.