Every AI word,
in plain English.

A

• A2AAgent2Agent protocol Agent2Agent: an open protocol that lets AI agents built by different companies discover…
• accelerationisme/acc in tech culture, the view that AI and technology should be developed as fast as possible…
• agent loop the repeating cycle at the heart of an agent: the model decides on an action, a tool…
• agent skillskill a packaged set of instructions, scripts, and resources that an AI agent can load when…
• agentic able to act independently toward a goal, taking multiple steps and making decisions with…
• agentic browserAI browser a web browser with a built-in AI agent that can read pages and act for you, such as filling…
• agentic coding software development in which an AI agent plans changes, edits many files, runs tests and…
• agentic commerce shopping in which AI agents search, compare, and buy products or services on a person's…
• agentic workflow a process in which one or more AI agents carry out a series of steps, such as research…
• AI agentagent an AI system that can pursue a goal over many steps on its own, deciding what to do next…
• AI assistant an AI product that helps a person with everyday tasks such as answering questions, writing…
• AI coding assistant a tool that helps programmers by suggesting code as they type, explaining code, writing…
• AI companion an AI character or chatbot designed for ongoing personal conversation, friendship, or…
• AI psychosisAI delusions (informal) a term used in the media for cases in which people develop strong false beliefs…
• AI safety the field concerned with making AI systems behave as intended and avoid causing harm, from…
• AI slopslop low-quality AI-generated content produced in bulk, such as spammy articles, strange images…
• AI washing exaggerating or inventing the use of AI in a product or company to attract customers or…
• AI winter a period when interest, funding, and progress in AI fall sharply after expectations are not…
• AI-first describing a company or strategy that treats AI as the default way to do work, checking…
• AI-native designed from the start around AI, rather than having AI features added to an existing…
• algorithm a set of step-by-step instructions for solving a problem or completing a task
• alignmentAI alignment the work of making an AI system's goals and behavior match what its designers and users…
• answer engine optimizationAEO optimizing content so it is chosen as the direct answer by voice assistants, featured…
• APIapplication programming interface application programming interface: a defined way for one program to send requests to…
• API key a secret code that identifies your account when your program calls an API, and that is used…
• artificial general intelligenceAGI a hypothetical AI that can understand, learn, and perform virtually any intellectual task a…
• artificial intelligenceAI the field of building computer systems that perform tasks normally thought to require human…
• artificial superintelligenceASI hypothetical AI that is far more capable than the best humans in nearly every field…
• attentionself-attention a technique that lets a model weigh how relevant every part of the input is to every other…
• autonomous vehicleself-driving car a car or other vehicle that can drive itself using cameras, sensors, and AI, with little or…

B

• background agentasync agent an AI agent that works on a task by itself in the cloud while you do something else, then…
• backpropagationbackprop the method used to train neural networks, which works backward from an error to calculate…
• base modelpretrained model a language model that has only been pre-trained, not yet tuned to follow instructions or…
• benchmark a public, standardized test used to compare AI models, such as a set of exam questions…
• biasalgorithmic bias systematic unfairness in an AI system's outputs, often learned from patterns in its…
• bitter lessonThe Bitter Lesson the observation that, over the long run, simple general methods that use more computation…
• black box a system whose inputs and outputs can be seen but whose internal reasoning cannot easily be…

C

• chain of thoughtCoT a series of intermediate reasoning steps a model writes out before its final answer, either…
• chatbot a computer program that holds a conversation with people through text or voice
• chunking splitting long documents into smaller pieces before creating embeddings, so a retrieval…
• citation in AI answers, a link or reference showing which source supports a statement
• clanker (slang, mocking) a robot or AI system, used jokingly or to express frustration with AI
• coding agent an AI agent built to work on software: it reads a codebase, writes and edits code across…
• cognitive debt the long-term cost to your own thinking skills that may build up when you hand writing…
• compute computing power, especially the chips and processing time used to train and run AI models
• computer usebrowser agent the ability of an AI agent to operate a computer or web browser like a person, by looking…
• computer visionCV the field of AI that enables computers to interpret images and video, for example…
• constitutional AI a training approach in which a model is guided by a written set of principles (a…
• context engineering the practice of deciding exactly what information, instructions, tools, and history go into…
• context rot the gradual decline in a model's accuracy and focus as its context window fills with more…
• context windowcontext length the maximum amount of text, measured in tokens, that a model can consider at one time…
• copilot an AI assistant built into a piece of software that suggests, drafts, or completes work…
• CUDA Nvidia's software platform for programming its GPUs, used by most AI frameworks

D

• dark factorysoftware factory a way of building software in which AI coding agents write, test, and ship code from…
• data centerAI data center a large building full of servers, chips, cooling, and power equipment; AI data centers can…
• data labelingannotation the work of tagging raw data with correct answers or categories, such as marking objects in…
• data void a topic with little reliable information online, so search engines and AI models fall back…
• dataset an organized collection of data used to train, test, or evaluate a model
• deep learning a type of machine learning that uses neural networks with many layers, allowing a system to…
• deep research an AI feature or agent that spends several minutes searching many sources, reading them…
• deepfake a realistic fake image, video, or audio recording of a real person, made with AI
• diffusion model a generative model that creates images (or audio or video) by starting from random noise…
• digital twin a detailed virtual copy of a real object, building, factory, or city, kept updated with…
• distillationknowledge distillation training a smaller "student" model to imitate the outputs of a larger "teacher" model…
• doomerAI doomer (often dismissive) a person who believes advanced AI is likely to cause catastrophe and…

E

• embedding a list of numbers (a vector) that represents the meaning of a piece of text, an image, or…
• embodied AIembodied intelligence AI that has a physical body, such as a robot or vehicle, and learns and acts by sensing and…
• emergent abilitiesemergence skills that seem to appear suddenly once a model passes a certain size, even though it was…
• epoch one complete pass through the entire training dataset
• evalevals a set of test cases used to measure how well a model or AI feature performs on the tasks…
• existential riskx-risk the possibility that advanced AI could cause human extinction or permanently and…

F

• few-shot done by including a handful of examples in the prompt so the model can copy the pattern…
• fine-tuning further training of an existing model on a smaller, focused dataset so it gets better at a…
• FLOPSFLOP floating-point operations (per second): the basic unit for counting how much computing work…
• foundation model a large model trained on broad data that can be adapted to many different tasks, serving as…
• frontier model one of the most capable AI models available at a given time, typically the newest and most…
• function calling an API feature in which a model, instead of replying in plain text, outputs a structured…

G

• generative AIGenAI AI that creates new content, such as text, images, music, video, or code, in response to a…
• generative engine optimizationGEO the practice of shaping web content so that AI chatbots and AI search features find it…
• glazingglaze (slang) excessive, undeserved praise; often used to complain that a chatbot flatters the…
• GPTgenerative pre-trained transformer short for generative pre-trained transformer: a language model that is first pre-trained on…
• GPUgraphics processing unit graphics processing unit: a chip originally designed for video games that can do many…
• gradient descent an optimization method that improves a model step by step by moving its parameters a little…
• grounding connecting a model's answer to trusted sources, such as documents, databases, or search…
• guardrails rules, filters, and checks placed around an AI system to keep its outputs and actions…

H

• hallucinationhallucinate a confident but false or made-up statement produced by an AI model, such as an invented…
• harness engineeringagent harness the work of building everything around an AI model that lets it act as a reliable agent…
• human-in-the-loopHITL describing a system in which a person reviews, approves, or corrects an AI's work at key…
• humanoid robothumanoid a robot with a roughly human body shape, including legs, arms, and often hands, designed to…
• hyperscaler one of the few giant cloud companies that operate data centers on a massive scale and rent…

I

• in-context learning a model's ability to pick up a new task from instructions or examples given in the prompt…
• inference the act of running a trained model to produce an output, such as answering a prompt or…
• instruction tuningsupervised fine-tuning fine-tuning a base model on examples of instructions paired with good responses, so it…
• interpretabilitymechanistic interpretability research into understanding what is happening inside a model, such as which internal…

J

• jagged frontier the uneven boundary of what AI can do well: it may handle a hard task brilliantly yet fail…
• jailbreak a prompt or trick designed to get an AI model to ignore its safety rules and produce…

K

• knowledge cutofftraining cutoff the date after which a model has no information from its training data
• KV cachekey-value cache memory that stores intermediate attention results for text a model has already processed…

L

• large language modelLLM a very large neural network trained on huge amounts of text to predict the next token…
• latency the delay between sending a request to a model and getting a response, often measured as…
• latent space the hidden, many-dimensional space of numbers in which a model represents concepts, where…
• LLM grooming deliberately flooding the internet with false or slanted content in the hope that AI models…
• LLM-as-a-judge using one language model to grade or compare the outputs of another, as a cheaper and…
• llms.txt a proposed plain-text file placed at the root of a website that gives AI models a short…
• LoRAlow-rank adaptation low-rank adaptation: a cheap fine-tuning method that freezes the original model and trains…
• loss functionloss a formula that measures how wrong a model's output is during training; training tries to…

M

• machine learningML a branch of AI in which computers learn patterns from data instead of following rules…
• MCPModel Context Protocol Model Context Protocol: an open standard for connecting AI applications to outside tools…
• memory a feature that lets an AI keep information about a user or task across separate…
• mixture of expertsMoE a model design in which many smaller sub-networks ("experts") exist, but only a few are…
• model the trained program that results from machine learning: a set of learned numbers plus the…
• model cardsystem card a document published alongside a model describing what it is for, how it was trained and…
• model collapse a gradual loss of quality and diversity that happens when models are trained on content…
• Moravec's paradox the observation that tasks humans find hard, like chess or math, are relatively easy for…
• multi-agent system a setup in which several AI agents, often with different roles, work together or check each…
• multimodal able to understand or produce more than one type of data, such as text, images, audio, and…

N

• narrow AIweak AI AI that is designed for, and good at, one specific task or a small set of related tasks…
• natural language processingNLP the field of AI concerned with getting computers to understand, generate, and work with…
• neural networkartificial neural network a computing system made of many simple connected units ("neurons") arranged in layers…
• neuralese the internal numerical 'language' a model might use to think and communicate, which humans…
• no-code describing tools that let people build apps, websites, or automations through visual…
• NPUneural processing unit neural processing unit: a small, power-efficient chip inside phones and laptops built to…

O

• on-device AIedge AI AI that runs directly on a phone, laptop, car, or other device instead of in the cloud…
• one-shot done with exactly one example included in the prompt
• opaque recurrencerecurrent depth a technique in which a model reasons by looping information through its internal layers…
• open weightsopen-weight model describing a model whose trained weights are published so anyone can download, run, and…
• open-source model a model released with its weights, code and, ideally, enough information about its training…
• orchestration the software layer that coordinates models, tools, agents, and data so a multi-step AI task…
• overfitting a problem in which a model learns its training examples too closely, including their noise…

P

• p(doom)p doom (informal) a person's estimated probability that AI will lead to catastrophic outcomes for…
• parameterparams one of the numbers inside a model that is adjusted during training; the total count of…
• parasocial describing a one-sided relationship in which a person feels close to someone or something…
• persona a role or character an AI is told to adopt, such as 'experienced editor' or 'friendly…
• physical AI AI that understands and acts in the physical world, powering robots, self-driving cars…
• post-training everything done to a model after pre-training to make it useful and safe, such as…
• pre-trainingpretraining the first and most expensive stage of building a model, in which it learns general patterns…
• prompt the text, question, or instruction a person gives to an AI model to get a response
• prompt caching reusing the already-processed beginning of a prompt across many requests, which cuts cost…
• prompt engineeringprompt engineer the practice of designing and refining prompts to get reliable, high-quality results from…
• prompt injection an attack in which hidden instructions are slipped into content an AI reads, such as a web…

Q

• quantization shrinking a model by storing its numbers with less precision (for example 4 bits instead of…

R

• RAMageddon (informal) the shortage and sharp price rise of computer memory chips caused by AI data…
• rate limit a cap on how many requests or tokens an account can send to an API in a given time
• reasoning in AI, a model's ability to solve problems that need several steps, such as math, planning…
• reasoning modelthinking model a language model trained to work through a problem step by step before giving its final…
• recursive self-improvementRSI a scenario in which an AI system improves its own design, and each improved version gets…
• red teamingred team deliberately attacking or stress-testing an AI system, often with outside experts, to find…
• reinforcement learningRL a way of training in which a model tries actions, receives rewards or penalties, and…
• reranking a second step in search where a more careful model re-scores the top results so the most…
• responsible AI the practice of designing, building, and deploying AI with attention to fairness…
• retrieval-augmented generationRAG a technique in which a system first searches a collection of documents for relevant…
• reward hackingspecification gaming when an AI finds a shortcut that earns a high score from its reward or test without…
• reward model a separate model trained to score how good a response is, used to guide reinforcement…
• RLHFreinforcement learning from human feedback reinforcement learning from human feedback: a training method in which people rate or…

S

• sampling the way a model picks each next token from its list of possible choices, which decides how…
• scaling laws observed patterns showing that model performance improves in a predictable way as you…
• SDKsoftware development kit software development kit: a ready-made library that makes it easier to use an API from a…
• semantic search search that matches the meaning of a query rather than its exact words, usually by…
• singularitytechnological singularity a hypothetical future point at which technological growth, driven by AI that improves…
• slopsquatting a security attack in which someone registers a software package under a name that AI coding…
• small language modelSLM a compact language model, usually with a few billion parameters or fewer, designed to run…
• sovereign AI a country's own AI capability, including data centers, models, and data, built so it does…
• speech-to-textSTT technology that converts spoken words into written text
• stochastic parrot a critical description of a language model as something that stitches together plausible…
• streaming sending a model's response token by token as it is generated, so text appears on screen…
• structured outputJSON mode model output that follows a fixed, machine-readable format such as JSON with specific…
• subagent an agent that a main agent creates or calls to handle one part of a task, usually with its…
• sycophancysycophantic an AI model's tendency to tell users what they want to hear, agreeing with them, praising…
• synthetic data training data generated by AI models or simulations rather than collected from the real…
• system promptsystem message hidden instructions given to a model before the conversation starts, which set its role…

T

• temperature a setting that controls how random a model's output is: low temperature gives safer, more…
• test-time computeinference-time compute extra computing power spent while answering a question, for example by letting the model…
• text-to-image describing AI that creates pictures from a written description
• text-to-speechTTS technology that turns written text into spoken audio, now often in highly natural or cloned…
• text-to-video describing AI that generates video clips from a written description
• throughput how much work an AI system can handle in a given time, often measured in tokens per second
• token the basic unit of text a language model reads and writes: a whole word, part of a word, a…
• tokenizer the component that splits text into tokens before it goes into a model, and joins tokens…
• tokenmaxxing the practice of pushing employees to use as many AI tokens as possible, treating token…
• tool use a model's ability to call external tools, such as a search engine, calculator, code runner…
• top-pnucleus sampling a sampling setting that lets a model choose only from the most likely next tokens whose…
• TPUtensor processing unit tensor processing unit: a chip designed by Google specifically for machine learning
• training the process of feeding data to a model and repeatedly adjusting its parameters so its…
• training data the collection of examples (text, images, code, audio, and so on) that a model learns from
• transformer a neural network design that processes a whole sequence at once and uses attention to work…
• Turing test a test of whether a machine can hold a text conversation so well that a person cannot…

V

• vector an ordered list of numbers; in AI, the form in which models represent words, images, and…
• vector databasevector store a database built to store embeddings and quickly find the ones most similar to a query…
• vibe codingvibe coder a way of building software by describing what you want to an AI in plain language and…
• vibe working office work done by describing the result you want to an AI agent inside a document…
• vision language modelVLM a model that takes both images and text as input and can describe, answer questions about…
• vision-language-action modelVLA a model that takes in camera images and spoken or written instructions and directly outputs…
• voice cloning using AI to copy a specific person's voice from a short recording so it can say new things
• VRAMvideo memory the memory built into a GPU; how much a model needs largely decides which graphics card can…

W

• watermarking hiding an invisible signal in AI-generated text, images, audio, or video so it can later be…
• weights the learned numbers stored in a trained neural network that determine how strongly each…
• worksloppromptslop AI-generated work, such as reports, memos, emails, or slides, that looks polished but lacks…
• world model an AI system that learns how the physical world behaves, such as how objects move and…
• wrapperGPT wrapper (often dismissive) a product that is mainly a user interface around someone else's AI…

Z

• zero-click search a search where the user gets the answer directly on the results page, increasingly from an…
• zero-shot done without giving the model any examples, relying only on instructions