Learn
AI, in plain English.
42 terms you meet on the Wire, explained without jargon, each with an example.
Basics
Large language modelA program that learned to use language by reading enormous amounts of text.TokenThe small chunk of text an AI reads and writes, often part of a word.Context windowHow much text an AI can keep in mind at one time.ParametersThe internal dials a model adjusts while learning, numbering in the billions.Training vs inferenceTraining is teaching the model once; inference is using it every day.PromptThe instruction or question you give an AI to get a response.System promptHidden setup instructions that shape how an AI behaves in a chat.HallucinationWhen an AI states something false with total confidence.MultimodalAn AI that can handle more than just text, like images, audio, or video.TemperatureA setting that controls how predictable or creative an AI's answers are.Knowledge cutoffThe date after which an AI simply has not read anything.
Models
Reasoning modelAn AI that pauses to think through a problem step by step before answering.Mixture of expertsA big model split into specialists, only a few of which handle each question.Open weights vs closedWhether a model's inner settings are shared publicly or kept private.DistillationTraining a small, cheap model to imitate a big, expensive one.QuantizationRounding a model's numbers to make it smaller and faster.Fine-tuningGiving an already-trained model extra practice on a specific task.EmbeddingsTurning words or images into numbers that capture their meaning.BenchmarkA standard test used to compare how good different AI models are.Frontier modelThe most advanced, cutting-edge AI models available at any moment.
Building
RAG (retrieval-augmented generation)Letting an AI look up real documents before it answers.AI agentAn AI that can take actions and complete multi-step tasks on its own.Tool useAn AI reaching outside itself to use calculators, search, or other software.MCP (Model Context Protocol)A common standard that lets AI plug into apps and data sources.Function callingHow an AI asks software to run a specific action with the right details.Vector databaseA special store that finds information by meaning rather than exact words.GuardrailsSafety checks that keep an AI's behavior within acceptable limits.EvalsTests that measure whether an AI system is actually doing its job well.Prompt injectionSneaky hidden instructions that trick an AI into misbehaving.
Safety
AlignmentMaking sure an AI's goals and behavior match what people actually want.Red-teamingDeliberately attacking an AI to find its weaknesses before others do.JailbreakA trick that gets an AI to break its own safety rules.WatermarkingA hidden signal marking content as AI-generated.Model cardsA short fact sheet describing what an AI model can and cannot do.Compute governanceOverseeing the powerful computer chips used to build advanced AI.AI safety instituteA government body that tests and studies the risks of advanced AI.
Business
APIA doorway that lets other software talk to an AI model.Tokens as costYou usually pay for AI by the token, for both what goes in and comes out.Rate limitsA cap on how many requests you can send an AI in a given time.On-device AIAI that runs right on your phone or laptop instead of a distant server.Sovereign AIA country building and controlling its own AI within its borders.Inference vs training costTraining is a giant one-time bill; inference is the ongoing running cost.