Terms to Know

Artificial Intelligence (AI)

Software designed to perform tasks that typically require human intelligence — recognizing patterns, understanding language, making decisions. The term covers everything from simple recommendation engines to large language models like the one generating this text.

Machine Learning (ML)

A subset of AI in which systems learn from data rather than following explicitly programmed rules. The more data they process, the better they get at the task they were trained on.

Neural Network

A computing architecture loosely modeled on the human brain — layers of interconnected nodes that process and pass information. The foundation of most modern AI systems.

Large Language Model (LLM)

An AI system trained on massive volumes of text that generates human-like language. GPT, Claude, and Gemini are all LLMs. They predict what word or phrase comes next based on patterns learned during training.

Training Data

The dataset an AI system learns from. Its quality, scope, and embedded biases directly determine what the AI knows, how it reasons, and where it fails.

Parameters

The internal numerical values that define how an AI model behaves. A model with more parameters can capture more complex patterns — but parameters are set by humans, not discovered independently by the AI.

Inference

What happens when an AI model is actually used — applying what it learned during training to respond to new input. Training is the education. Inference is the exam.

Hallucination

When an AI generates confident, plausible-sounding content that is factually wrong or entirely fabricated. Not a bug being fixed — a structural consequence of how language models work.

Confabulation

A more precise term for hallucination. Borrowed from neuroscience — it means filling memory gaps with invented content that feels real. AI confabulates for the same structural reason humans sometimes do: the system is optimized for coherence, not accuracy.

Bias

Systematic errors in AI output that reflect prejudices, gaps, or inequities in the training data. AI doesn't inherit human wisdom — it inherits human patterns, including the ones we're not proud of.

Generative AI

AI systems that produce new content — text, images, audio, video, code — rather than simply classifying or analyzing existing content. LLMs are one category of generative AI.

Narrow AI

AI designed to perform one specific task — playing chess, recognizing faces, translating languages. All current AI is narrow AI, despite how capable it appears.

General AI (AGI)

A hypothetical AI that could perform any intellectual task a human can. Does not currently exist. Significant debate exists about whether it is achievable and what it would mean.

Natural Language Processing (NLP)

The field of AI focused on enabling machines to understand and generate human language. The foundation of chatbots, translation tools, and search engines.

Computer Vision

AI that interprets and understands visual information — images, video, medical scans. Powers facial recognition, autonomous vehicles, and cancer screening tools.

Fine-tuning

Taking a pre-trained AI model and further training it on a specific dataset to specialize its behavior. Like giving a generalist employee focused on-the-job training.

Prompt

The input a user gives an AI system — a question, instruction, or context. The quality of the prompt significantly affects the quality of the response.

Context Window

The amount of text an AI can process at one time — its working memory for a single session. Everything outside the context window is invisible to the model.

RAG (Retrieval-Augmented Generation)

A technique that allows AI to search external documents or databases before generating a response — improving accuracy by grounding answers in retrieved information rather than training data alone.

Token

The basic unit AI models process — roughly a word or word fragment. Models have token limits per session, which is why very long conversations can get cut off or degrade.

---