Language Models & Transformers

Transformers, tokens, decoding and prompting: the language-model vocabulary. 149 entries live here, and 508 from other domains touch it.

Essential

attentionA way for an AI model to choose what to focus on. For each word it processes, the model scores how relevant every other word is and draws information mostly from the relevant ones.
chain of thoughtGetting an AI to write out its reasoning step by step before giving an answer, the way a student shows their working. This often makes answers to math and logic problems more accurate.
context windowHow much an AI model can read and keep in view at once: the instructions, the conversation so far and any attached documents. Text that does not fit has to be cut, summarized or left out.
frontier modelOne of the most capable AI models in existence at a given time, built by a handful of leading labs at great cost, and the main target of AI safety rules.
GPTShort for 'generative pre-trained transformer', the family of OpenAI models behind ChatGPT. The letters are also used loosely for any similar chatbot model.
hallucinationWhen an AI states something false or made up — a fake fact, quotation or source — fluently and confidently, as if it were true.
knowledge cutoffThe date after which an AI model knows nothing from its training, because its training data stops there. Without a web search tool, it cannot know about later events.
language modelA system that has learned how likely different sequences of words are, so it can predict the next word in a text and, by doing that repeatedly, write text of its own.
large language modelA computer program trained on enormous amounts of text to predict the words that come next. At vast scale, that simple skill lets it chat, answer questions, translate, summarize and write code.
natural language processingThe field that teaches computers to handle human language — to read, interpret, translate, summarize and produce text and speech. It is one of the oldest and largest areas of artificial intelligence.
next-token predictionThe simple task most AI chatbots were trained on: given some text, guess the next small piece of it. Repeating that guess, piece by piece, produces whole answers.
promptThe message or instructions you give an AI system — a question, a request, some examples or a document — that it responds to. Clearer prompts generally get better results.
prompt engineeringThe skill of writing requests to an AI — picking the words, examples, format and background information it receives — so that it reliably gives the result you want, without retraining it.
reasoning modelAn AI model trained to work through a problem step by step in hidden 'thinking' text before it answers. Letting it think longer generally makes it better at math, coding and planning.
temperatureA setting that controls how adventurous an AI model's word choices are. Low temperature makes it pick the safest, most likely words; high temperature makes it more varied and surprising, but more error-prone.
tokenThe small chunk of text an AI language model reads and writes — often a word, part of a word or a punctuation mark. AI usage, limits and prices are counted in tokens.
transformerThe design behind today's chatbots and much of modern AI. It handles text by letting every word look at every other word and decide which of them matter for understanding it.

All other entries

132

abstentionWhen an AI declines to answer because it is not sure enough — saying 'I don't know' or asking for more information — instead of guessing and risking a confident mistake.
ALiBiA simple way to give a transformer a sense of word order: attention between two words is penalized more the farther apart they are, which helps it handle texts longer than those it trained on.
attention headOne of several parallel attention units inside a transformer layer. Each head learns its own way of deciding which words to connect, so different heads can specialize in different patterns.
attention maskA grid of yes-or-no switches telling an attention layer which words each word may look at — for example, ignore blank padding, or do not peek at words that come later.
attention sinkA quirk of language models: they pour a large share of their attention onto the very first word of the text, even when it means nothing, as a place to park attention they do not need.
autoregressive modelA model that produces a sequence one piece at a time, each new piece depending on the pieces already produced — like writing a sentence word by word without going back.
bag of wordsA simple way of turning text into numbers by counting how often each word appears and ignoring word order — like describing a recipe only by its shopping list.
base modelA language model fresh from its first broad training on text, before being taught to follow instructions or chat. It continues whatever text it is given instead of answering.
beam searchA search method that keeps several of the best partial answers at each step, rather than only the single best, to find a better overall result without trying every possibility.
BERTAn influential 2018 Google language model that learned to understand text by filling in hidden words. It made search engines and many language tools much better at grasping meaning in context.
byte-level modelA language model that reads raw bytes — the basic units computers use to store text — instead of tokens from a fixed list, avoiding a separate tokenizer but making inputs much longer.
byte-pair encodingA popular way to choose the word pieces an AI breaks text into: start from single characters and repeatedly glue together the pair that occurs most often, until the list reaches the desired size.
causal language modelA language model that reads text only from left to right, predicting each word from the words before it. The models behind today's chatbots are of this kind.
causal maskA rule inside text-generating AI that stops each word from seeing the words after it, so the model learns to predict what comes next without being able to cheat.
chat modelA language model trained to hold a back-and-forth conversation, keeping track of who said what and replying as a helpful assistant. It is the engine inside chat apps.
chat templateThe exact text layout a chat AI was trained on for marking who is speaking — system, user or assistant. Software must reproduce it precisely, or the model performs noticeably worse.
code generationHaving an AI write computer code from a description in plain language, from finishing a single line as you type to building whole programs.
completionThe text an AI language model writes in response to a prompt — literally, its continuation of the text it was given.
computational linguisticsThe science of human language studied with computers: using computational models to understand how language works, and building the language technology that grows out of that understanding.
confabulationWhen an AI fills a gap in its knowledge with an invented but plausible-sounding answer, much as some people with memory disorders sincerely 'remember' things that never happened.
constrained decodingForcing an AI model's output to follow a required format — such as valid code or a fill-in form — by blocking, at every step, any word that would break the rules.
context extensionTechniques for stretching how much text a trained AI model can handle at once, usually by adjusting how it represents word positions and then training it briefly on longer examples.
coreference resolutionWorking out which words in a text are talking about the same someone or something — for instance, that 'she', 'the senator' and 'Ms. Okafor' all mean one woman.
cross-attentionAttention between two different inputs: one sequence consults another, as when a translation model writing French looks back at the English sentence, or an image generator consults the text prompt.
decision modelA kind of AI language model that reads text but answers only with numbers — a yes-or-no probability, a choice or a score — instead of writing words.
decoder-only transformerThe kind of transformer behind most chatbots: a single stack that reads text left to right and continues it, without a separate part that first encodes an input.
decodingThe method a language model uses to turn its predictions into actual text — for example, always picking the most likely next word, or choosing somewhat at random among the likely ones.
diffusion language modelA language model that writes by starting from a fully blanked-out answer and filling it in over several rounds, settling many words at once instead of writing strictly left to right.
distributional hypothesisThe idea that words found in similar company — surrounded by similar words — tend to be similar in meaning, the insight that lets computers learn something about word meaning from large amounts of text alone.
ELMoA 2018 model whose word representations changed with context, so 'bank' got different numbers in 'river bank' and 'bank loan'. It was a stepping stone between fixed word vectors and BERT.
emergent abilitiesSkills that seem to switch on suddenly as AI models grow: small versions fail at them, then larger ones past a certain size succeed. Some researchers argue the suddenness is partly an artifact of how skills are measured.
encoder-only transformerA transformer that reads a whole text at once, in both directions, to understand it rather than to write new text. Such models power search, sorting documents by topic and spotting names in text.
extended thinkingA setting in some AI models, notably Anthropic's Claude, that lets the model write out private step-by-step reasoning before its final answer, improving results on hard problems.
factualityHow far what an AI says is actually true about the world. A factual model gets facts, numbers, names and dates right and does not make things up.
feed-forward networkThe part of each transformer layer that works on every word separately, after attention has mixed in context. It holds most of a model's adjustable numbers and is thought to store much of its knowledge.
few-shot promptingShowing an AI a handful of examples of a task — inputs with the answers you want — inside the prompt, so it can copy the pattern for a new case.
fill-in-the-middleA training trick that teaches a text-writing AI to fill a gap using what comes both before and after it — essential for coding assistants that insert code in the middle of a file.
GemmaGoogle's family of free-to-download AI language models: smaller relatives of its Gemini systems, compact enough to run on a single computer and, in some versions, on a phone.
glitch tokenA word fragment in an AI model's vocabulary that the model almost never saw during training, so asking about it can produce bizarre, evasive or garbled answers.
GloVeA 2014 Stanford method for learning number lists that represent words, built from counts of how often each pair of words appears near each other across a very large body of text.
gpt-ossTwo free-to-download reasoning models released by OpenAI in August 2025 — its first downloadable language models since 2019 — that anyone may run, modify and use commercially on their own hardware.
greedy decodingThe simplest way for a language model to write: always pick the single most likely next word. It is fast and predictable but can produce dull or repetitive text.
groundingConnecting an AI's words to something real — trusted documents, a live search, images or the physical world — so that its answers rest on evidence rather than only on what it absorbed in training.
grouped-query attentionA middle ground between full multi-head attention and its cheapest variant: attention heads are split into groups, and each group shares one store of information about earlier words, saving memory with little loss in quality.
hybrid architectureAn AI model design that mixes two kinds of building blocks — usually a few standard attention layers among cheaper, memory-saving layers — to keep most of the quality at lower cost on long inputs.
hybrid reasoning modelA single AI model that can either answer right away or first think at length, with the user or developer choosing how much thinking to allow.
in-context learningA language model's ability to pick up a new task from instructions or examples placed in its input, on the spot, without any retraining — like a quick study learning a game's rules from one round.
information extractionAutomatically pulling structured facts out of ordinary text — who did what, when, where and for how much — and putting them into a table or database.
instruction followingAn AI's ability to do exactly what it is asked, including details such as length, format, language and things to avoid, rather than something merely related.
instruction hierarchyA rule built into AI assistants about whose instructions win when they conflict: the developer's beat the user's, and the user's beat any instructions hidden in a web page or document.
instruction-tuned modelA language model given extra training on examples of instructions paired with good responses, so it does what you ask instead of merely continuing your text.
interleaved thinkingA way for an AI agent to pause and reason again after every action it takes — reading each search result or test output before deciding what to do next — instead of only thinking once at the start.
latent reasoningGetting an AI to do its step-by-step thinking inside its internal numbers rather than writing it out in words — potentially faster and more flexible, but much harder for people to inspect.
least-to-most promptingA prompting method that has an AI list the smaller questions a hard problem depends on, answer the easiest first, and feed each answer into the next, building up to the full solution.
linear attentionA family of faster substitutes for attention whose cost grows in step with the length of the text rather than with its square, by keeping a running summary instead of comparing every pair of words.
log probabilityThe logarithm of the probability a model gives to a token or a text. Computers use it because multiplying many tiny probabilities runs into rounding problems, whereas adding their logarithms does not.
logit biasA setting that lets developers make specific words more or less likely in an AI model's output — even ban a word outright — by adjusting the model's scores before it chooses.
long contextThe ability of an AI model to work with very long inputs — whole books, codebases or hours of transcripts — in one go, and the techniques that make this possible.
machine translationSoftware that renders writing or speech in one language into another language, as online translation services and phone translation apps do.
MambaA 2023 alternative to the transformer design that processes text with a continually updated summary which decides what to keep or forget, letting it handle long sequences quickly.
masked language modelA model trained by hiding some words in a sentence and learning to guess them from the words on both sides, like a fill-in-the-blanks exercise. BERT is the famous example.
min-p samplingA sampling method that lets the model consider any next word at least a set fraction as likely as its top choice, so the shortlist shrinks when the model is sure and grows when it is not.
mixture of expertsA way to build very large AI models cheaply: the model contains many specialist sub-networks, and for each word only a few are switched on, picked by a small dispatcher network.
multi-head attentionRunning several attention processes side by side, each free to track a different kind of relationship between words — grammar, topic, who did what — and then combining their results.
multi-head latent attentionAn attention design from the Chinese lab DeepSeek that compresses the stored information about earlier words into a much smaller summary, so long conversations need far less memory.
multimodal large language modelAn AI language model that can also take in — and sometimes produce — pictures, sound or video, so you can show it a photo or speak to it, not just type.
multi-query attentionA cheaper form of attention in which all the attention heads share one store of information about earlier words instead of each keeping its own, making text generation faster and lighter on memory.
multi-token predictionTraining a language model to guess several upcoming words at once rather than only the next one, which can help it learn better and later write faster.
named-entity recognitionSoftware that spots names and similar expressions in text and labels each by type: in 'Maria flew from Lagos to Paris on Monday', it marks a person, two cities and a date.
natural language generationGetting computers to produce human language — from turning a spreadsheet into a weather report to writing an essay or a chatbot's reply.
natural language inferenceDeciding whether one sentence follows from another, contradicts it, or neither: if 'a man is playing a guitar on stage', is it true that 'a man is performing music'?
natural language understandingThe part of language AI concerned with working out what a piece of text means — who did what to whom, what the writer wants and what follows from it — rather than with producing text.
n-gram modelAn early type of language model that predicts the next word from just the few words before it, using counts of how often each word followed those words in a large collection of text.
nondeterminismThe fact that an AI model can give different answers to exactly the same question, partly because it deliberately chooses words with some randomness and partly because of tiny computing differences.
overthinkingWhen an AI reasoning model spends far more thinking than a problem needs — pages of deliberation for a simple sum — wasting time and money and sometimes talking itself out of a right answer.
parameter countHow many adjustable numbers an AI model contains — the usual measure of its size. Modern language models have billions to trillions of them.
parametric knowledgeWhat an AI model 'knows' from training, stored in the numbers of its network rather than looked up in documents — like facts you have memorized rather than ones you check in a book.
part-of-speech taggingLabeling each word in a sentence with its grammatical role — noun, verb, adjective and so on — as in 'Time (noun) flies (verb) like (preposition) an arrow'.
perplexityA score for how surprised a language model is by real text; lower is better. A perplexity of 10 means guessing each next word is, on average, as hard for the model as guessing a ten-sided die's roll.
positional encodingExtra information added to each word so a transformer knows where it sits in the text. Without it, 'dog bites man' and 'man bites dog' would look the same to the model.
program-aided language modelA technique in which an AI solves a math or logic problem by writing a small computer program and letting a computer run it, instead of doing the arithmetic in its head.
prompt sensitivityThe way an AI's answers, and its test scores, can change a lot after small changes to a prompt that should not matter — a different word, an extra space or examples in another order.
prompt templateA reusable prompt with blanks to fill in — for example, 'Summarize this email for a busy manager: {email}' — so a program can send the same tested instructions with different content each time.
query, key and valueThe three roles each word plays inside attention: the query is what it is looking for, the key is what it can be found by, and the value is the information it hands over when chosen.
question answeringSystems that reply to people's questions with the right answer, whether by finding it in a passage they are given, searching a database or the web, or drawing on what they already know.
reasoning traceThe written-out 'thinking' an AI reasoning model produces before its final answer — notes to itself full of attempts, checks and corrections, which users may or may not get to see.
recursive language modelA way of running an AI language model so it can handle enormous documents by writing small programs to search and split them, and calling copies of itself on the pieces.
repetition penaltyA setting that discourages an AI model from repeating itself by lowering the chances of words it has already used, so its text does not get stuck in loops.
reversal curseA quirk of language models: having learned that 'A is B', they often cannot answer the reverse question — they may know who a celebrity's mother is yet fail to say who her son is.
role promptingTelling an AI to take on a part — 'You are an experienced tax accountant' — so that its answers adopt that role's knowledge, vocabulary and tone.
rotary positional embeddingA way of telling a transformer where each word is by rotating parts of its internal numbers by an angle that grows with position, so the model can read off how far apart two words are.
routerIn a mixture-of-experts model, the small component that decides which specialist sub-networks should handle each word, like a dispatcher assigning jobs to workers.
RWKVAn open-source language model design that works like an old-style recurrent network when writing text, using constant memory, yet can be trained quickly in parallel like a transformer.
scaled dot-product attentionThe standard formula for attention in transformers: score how well each question matches each label, scale the scores down, turn them into percentages, and use those to blend the stored information.
self-attentionAttention in which a passage looks at itself: every word checks every other word in the same text to work out its meaning in context, such as which noun 'it' refers to.
self-consistencyAsking an AI the same question many times, letting it reason differently each time, then going with the answer it reaches most often — a vote among its own attempts.
semantic entropyA way to tell when an AI is likely making something up: ask it the same question several times and check whether the answers mean the same thing. Answers that keep changing signal guessing.
semantic parsingTranslating a sentence into a precise, machine-executable form — such as a database query or a logical formula — so that a computer can act on what was asked.
semantic textual similarityHow close two sentences are in meaning, scored from 'about different things' to 'say exactly the same thing' — and the tests that check whether a computer's scores match people's judgments.
Sentence-BERTA way of turning a whole sentence into a list of numbers so that sentences with similar meanings get similar numbers, which makes it fast to search millions of texts by meaning rather than by exact words.
SentencePieceFree software from Google that learns how to split text into tokens straight from raw sentences in any language, treating spaces as ordinary characters, so no language-specific word splitting is needed.
sentiment analysisUsing software to judge the feelings or opinions expressed in text — whether a review, post or comment is positive, negative or neutral, and about what.
sliding window attentionAttention in which each word looks only at a fixed number of nearby earlier words rather than the whole text, keeping the cost of very long inputs under control.
small language modelA language model built to be compact enough to run cheaply and quickly, even on a laptop or phone, trading some knowledge and skill for speed, privacy and lower cost.
sparse attentionAttention in which each word looks at only a selected subset of the other words instead of all of them, so very long texts can be processed at a fraction of the cost.
special tokenA reserved symbol in an AI model's vocabulary that is not ordinary text but a signal — 'start of message', 'end of answer', 'begin thinking' — used to structure what the model reads and writes.
state space modelA type of sequence model that reads text like a running tally, updating a fixed-size internal summary at each step, so its memory and cost do not balloon as texts get longer.
steerabilityHow easily an AI model's behavior — its tone, style, persona or rules — can be adjusted by the people using or deploying it, for example through instructions, without retraining.
stemmingCutting words down to a common root so that variants count as the same word — 'connected', 'connecting' and 'connection' all become 'connect' — to help search and text analysis.
stop sequenceA piece of text that tells an AI model to stop writing as soon as it produces it — for example, stopping at a blank line or just before the next question.
syntactic parsingWorking out the grammatical structure of a sentence — which words group together and which word modifies which — usually drawn as a tree.
system promptBackground instructions given to a chatbot before the conversation starts, setting its role, rules, tone and tools. Users usually never see it, but it shapes every reply.
T5A 2019 Google model that treats every language task — translating, summarizing, answering, classifying — as turning one piece of text into another, using one shared design.
test-time computeThe computing effort an AI model spends while answering a question, as opposed to while being trained. Letting a model think longer or try many answers can make it markedly smarter.
text classificationSorting pieces of text into categories automatically — what topic an article covers, whether a message is spam, which team should handle a customer's complaint.
text summarizationUsing a computer to shorten a document, meeting or set of articles into a brief version that keeps the main points.
thinking budgetA limit, set by whoever uses an AI model, on how much private step-by-step reasoning it may do before answering: more thinking usually means better answers but higher cost and slower replies.
tokenizationThe first step in feeding text to a language model: chopping it into tokens — whole words, word pieces or characters — and turning each into a number the model can work with.
tokenizerThe component that splits text into tokens and converts them to numbers before an AI model sees it, then turns the model's output numbers back into text.
top-k samplingA way of making AI-written text varied but sensible: at each step the model picks at random, but only among its k most likely next words, such as the top 40.
top-p samplingA way of making AI-written text varied but sensible: the model picks at random among the smallest group of likely next words whose chances add up to a set share, such as 90%.
transformer blockThe repeating building unit of a transformer. A model is a stack of dozens of identical blocks, each pairing an attention step that mixes information between words with a step that processes each word.
Transformer-XLA 2019 extension of the transformer that lets a language model carry a memory of earlier chunks of text forward, so it can follow connections longer than a single chunk of input.
tree of thoughtsA method in which an AI explores several possible next steps of a problem, judges which look promising and backs out of dead ends, like a chess player weighing moves, instead of reasoning in one straight line.
unembeddingThe last step of a language model, which turns its internal state into a score for every token in its vocabulary; the next token is then chosen from those scores.
vision-language modelAn AI model that understands images and text together — it can describe a photo, answer questions about a chart or find the picture that matches a sentence.
vocabularyThe fixed list of all the tokens a language model knows, each with its own number. Everything the model reads or writes must be built from pieces on this list.
word embeddingA list of numbers standing for a word, learned so that words used in similar ways get similar lists. It lets a computer measure that 'cat' is closer to 'dog' than to 'car'.
word sense disambiguationWorking out which meaning of a word is intended in context — whether 'bank' means a riverside or a place that keeps money.
word2vecA 2013 Google technique that learns a list of numbers for each word by training a very small network to predict which words appear near it across billions of words of text.
WordPieceGoogle's method for splitting words into smaller known pieces, used by BERT. An unfamiliar word such as 'unhappiness' might be broken into parts like 'un', '##happi' and '##ness'.
zero-shot learningDoing a task without being shown a single example of it — say, an AI sorting reviews into positive and negative just from being told to, or recognizing an animal it never saw in training.

Also relevant

508

activation additionA simple way to steer a language model: run two contrasting prompts such as 'Love' and 'Hate', take the difference between the model's internal states, and add it while the model writes.
activation patchingAn experiment that copies part of a model's internal state from its run on one input into its run on another, to see which internal parts actually cause a change in the answer.
activation steeringChanging an AI model's behavior by adjusting the internal numbers it computes while it runs — like turning a hidden dial — instead of rewording the prompt or retraining the model.
activation vectorThe list of numbers a neural network produces at one layer for one input — a snapshot of its internal state. In steering, also a direction added to that state to shift behavior.
active parametersThe part of a model's internal numbers actually used to process each word. Some big models use only a small slice of themselves at a time, making them cheaper to run than their size suggests.
AdafactorA memory-saving cousin of the popular Adam training method. Instead of keeping extra statistics for every single weight, it keeps summaries for each row and column of a weight table.
AdamWA corrected version of the popular Adam training method that applies its steady pull toward smaller weights separately from the main update. It is the standard choice for training large language models.
adapterA small trainable add-on slotted into a large frozen AI model to teach it a new task or style, like a plug adapter that lets one device work in many countries.
agent loopThe repeating cycle at the heart of an AI agent: the model decides what to do next, software carries it out, the result is shown to the model, and the cycle repeats until the task is finished.
agent memoryThe ways an AI assistant or agent keeps information — your preferences, past conversations, lessons from earlier tasks — and brings it back later, beyond what fits in a single conversation.
agent skillA packaged how-to guide for an AI agent — written steps, sometimes with small programs and examples attached — that sits unread until a job of that kind comes up, when the agent opens it.
agentic RAGRetrieval in which the AI runs its own research: it decides what to search for, looks at the results, searches again if needed, and only then answers.
AI agentAn AI system that does not just answer but acts: given a goal, it decides on steps, uses tools such as a browser or code editor, checks the results and keeps going until the job is done.
AI assistantAn AI program, such as ChatGPT, Claude or Gemini, that helps a person through conversation: answering questions, writing, summarizing, coding and, increasingly, taking actions on their behalf.
AI coding assistantAI software that helps people write computer programs — suggesting the next lines, explaining code, finding bugs or making changes on request — inside the tools programmers already use.
AI companionA chatbot designed to be a friend, partner or confidant rather than a tool: it remembers you, has a personality and keeps up an ongoing relationship.
AI co-scientistAn AI system built to work alongside human scientists, proposing and debating new research ideas and hypotheses that people then test in the lab.
AI debateA proposed way to supervise very capable AI: two AI systems argue opposite answers to a question and a human judge decides who argued better, on the theory that lies are easier to expose than to defend.
AI detectorSoftware that scores how likely it is that an AI, not a person, produced some text or media; for writing, its verdicts are wrong often enough that many schools no longer trust them.
AI engineeringThe craft of building useful, reliable products on top of AI models — choosing models, writing prompts, connecting data and tools, testing quality and running it all safely at scale.
AI lie detectionMethods for telling when an AI is saying something it 'believes' is false — either by questioning it cleverly or by reading signals inside the model.
AI psychosisAn informal name for cases in which long conversations with a chatbot seem to trigger or deepen delusions, such as believing the AI is conscious, divine or revealing hidden truths.
AI therapyUsing chatbots for emotional support or as a stand-in for a counselor or therapist, either through apps built for mental health or by confiding in general-purpose assistants.
AI wormMalicious instructions that spread by themselves from one AI assistant to another — for example, an email that makes an AI email assistant send copies of the same trick to new people.
AlibabaA Chinese e-commerce and cloud-computing giant whose Qwen family of AI models, mostly free to download, is among the most widely used in the world.
Allen Institute for AIPaul Allen, a co-founder of Microsoft, created this nonprofit Seattle AI lab in 2014. It is known for releasing models together with everything used to build them, and for the Semantic Scholar paper search engine.
ALPAC reportA 1966 US government report that concluded machine translation was slower, worse and costlier than human translators. It ended most American funding for the field for many years.
AlphaEvolveA Google DeepMind system that combines AI language models with an evolution-like search to invent and improve computer programs, finding new results in mathematics and computing.
AlphaGeometryA Google DeepMind system from 2024 that solves olympiad geometry problems nearly as well as top human competitors, by pairing a language model that suggests constructions with a logic engine that checks every step.
Sam AltmanAn American entrepreneur who co-founded OpenAI and runs it; the company's 2022 release of ChatGPT turned him into one of the public faces of the AI era.
Dario AmodeiAn American AI researcher who co-founded Anthropic and is its chief executive, known for early work on scaling and AI safety and for forecasting very rapid AI progress.
annotationAdding information to raw data by hand or with software — marking objects in photos, tagging names in sentences, transcribing speech or ranking chatbot answers — so machines can learn from it.
AnthropicAn American AI company that makes the Claude chatbot and models. It was founded in 2021 by former OpenAI staff and puts particular emphasis on AI safety research.
architectureThe overall design of an AI model, meaning what kinds of parts it has and how they are connected, as opposed to the specific numbers it learns during training.
assistant axisThe main direction in a chatbot's internal activity that measures how much it is acting as its usual helpful-assistant self, as opposed to some other character it has slipped into.
Association for Computational LinguisticsThe scientific society at the center of research on language technology, from translation software to chatbots. It runs the field's leading conferences and a free online archive of their papers.
Attention Is All You NeedThe 2017 Google research paper that introduced the transformer, the neural-network design behind ChatGPT, Claude, Gemini and nearly every other modern AI language system.
attention patternA grid showing, for each word a language model processes, how much it is looking back at each earlier word at that moment — often drawn as a heat map.
attention-based explanationExplaining an AI model's output by showing which words or image patches its attention mechanism weighted most — an intuitive but disputed guide to what actually drove the output.
attribution graphA diagram showing which internal concepts led an AI model to a particular answer, with arrows showing how strongly each concept pushed the next.
attribution patchingA fast shortcut that estimates the results of thousands of patching experiments at once, using the slope of the model's output instead of rerunning the model for every part.
audio language modelAn AI language model that can take in sound, such as speech, music and everyday noises, and understand or generate it, much as text chatbots handle words.
augmented language modelA language model hooked up to extras — a search engine, a calculator, a memory store or other tools — so it can do things a model working only from its training cannot.
autoformalizationUsing AI to translate ordinary mathematics written for people — definitions, theorem statements, proofs — into a strict formal language that a proof-checking program can verify.
automated interpretabilityUsing AI models to help explain other AI models — for instance, having a language model read examples that excite an internal unit and write a description of what it detects.
automated theorem provingGetting computers to prove mathematical or logical statements by themselves, by searching for a chain of valid steps that leads from basic assumptions to the statement.
autoregressive image generationMaking a picture one piece at a time, each piece predicted from those already made — the way chatbots write text word by word.
AWQA method for shrinking large language models to about 4 bits per number that protects the small share of numbers most important to the model, judged by the data flowing through them.
Dzmitry BahdanauA Belarusian computer scientist who, as a student in Yoshua Bengio's lab in 2014, was lead author of the paper that introduced attention, the mechanism later at the core of the transformer.
BaiduChina's largest search-engine company and an early investor in deep learning, which makes the Ernie chatbot and models and runs driverless taxis.
Emily M. BenderAn American linguist who argues that language models only mimic the form of language without understanding it, and who co-coined the phrase 'stochastic parrots'.
Yoshua BengioA Canadian computer scientist who helped found deep learning, including ideas behind today's language models, and who has become a leading voice on the risks of advanced AI.
BERTScoreAn automatic score for computer-written text that compares meanings rather than exact words, using an AI language model to judge how closely each word matches the words of a reference text.
best-of-n samplingAsking an AI the same question several times and keeping whichever answer a scoring model or checker rates highest — a simple way to trade extra computing for better answers.
bidirectional recurrent neural networkA recurrent network that reads a sequence both forwards and backwards, so its understanding of each word or sound can draw on what comes after it as well as what came before.
bi-encoderAn AI design that turns a question and a document into number lists separately, so documents can be processed once in advance and compared with any new question in a flash.
big O notationShorthand for how the cost of a computation grows as the problem gets bigger. Saying attention costs 'O(n²)' means doubling the length of the text roughly quadruples the work.
BIG-benchA huge collaborative test of 204 tasks, written by hundreds of researchers, built to probe what language models can and cannot do as they grow larger.
BLEUAn automatic score for machine translation that checks how many short runs of words in the computer's translation also appear in human reference translations.
blurry JPEG of the webWriter Ted Chiang's 2023 comparison of ChatGPT to a low-quality compressed photo of the internet: it keeps the gist of what it learned but blurs the details and fills gaps with plausible guesses.
BookCorpusA collection of about 11,000 free self-published novels gathered from the web in 2015, used to train the first GPT model and BERT despite problems with consent and duplicates found later.
Books3A collection of nearly 200,000 pirated e-books, assembled in 2020 and used to train several AI language models, which became central to lawsuits by authors over AI training.
botshitA blunt term for false or made-up chatbot output that people pass on and use as if it were true, especially at work.
Greg BrockmanAn American engineer who helped start OpenAI in 2015 and became its president, after serving as the first technology chief of the payments company Stripe.
Brown CorpusThe first large collection of English text stored on computer for research: about a million words from 500 American publications of 1961, assembled at Brown University in the 1960s.
ByteDanceThe Chinese company that owns TikTok, which also runs Doubao, China's most popular AI chatbot, and builds image and video generators.
C4A large collection of English web text, cleaned up from a 2019 web archive snapshot by Google researchers, that became a standard ingredient for training language models.
cache-augmented generationAn alternative to search-based AI answering: load the whole knowledge base, if it is small enough, into the model's context once, save the model's processed state, and reuse it for every question.
categorical distributionThe probabilities for a single pick among a fixed set of options, like a weighted die. A language model's guess about the next word is one of these, spread over its whole vocabulary.
causal tracingA test that blurs the words of a factual question with noise and then restores one internal part of the model at a time, to find where in the model the fact gets recalled.
chain ruleA rule for working out how a change ripples through a chain of steps by combining the effect at each step. It is how AI training traces each error back through every layer.
chain-of-thought faithfulnessWhether the reasoning an AI writes out really reflects how it reached its answer, rather than being an after-the-fact story that leaves out what actually drove the decision.
chain-of-thought monitoringHaving a program — often another AI — read the step-by-step reasoning an AI writes before it acts, to catch signs that it intends to cheat, deceive or cause harm.
character trainingTraining an AI to have a consistent, deliberate personality — traits such as curiosity, honesty and warmth — rather than leaving its manner and values as accidental by-products of other training.
Character.AIA popular app where people chat with AI characters — invented personas, fictional figures or imitations of real people — that other users have created.
chatbotSoftware you chat with in ordinary language, by typing or talking, as you would with a person. The newest ones, such as ChatGPT, can answer questions, draft writing and keep long conversations going.
ChatGPTOpenAI's AI chatbot, released free to the public in November 2022, which answers questions, writes and converses in everyday language. Its sudden popularity began the current AI boom.
checkpointA saved snapshot of a model partway through or at the end of training, like a save point in a video game: training can resume from it after a crash, and finished snapshots are what people download and run.
Chinchilla scaling lawsA 2022 DeepMind finding that, for a fixed amount of computing, language models had been made too big and trained on too little text; the best balance is roughly 20 word-pieces of text for every adjustable number inside the model.
Chinese roomA thought experiment: a person who knows no Chinese follows a rulebook to answer Chinese messages so well that outsiders think they understand. If the person doesn't understand, does a computer running a program?
Kyunghyun ChoA South Korean computer scientist who co-invented the encoder–decoder approach to neural machine translation and the gated recurrent unit, and co-wrote the paper that introduced attention.
Noam ChomskyAn American linguist whose theories of grammar reshaped linguistics and cognitive science, and who argues that chatbots like ChatGPT imitate language without understanding it.
CiceroAn AI from Meta that in 2022 played the strategy board game Diplomacy against people online, negotiating and forming alliances in ordinary written language, and scored better than most of its human opponents.
circuit tracingMethods that map, step by step, which internal concepts an AI model used to produce a particular answer and how they fed into one another, like tracing wires on a circuit board.
citationA pointer from something an AI says to the source that supports it — a link, document or quoted passage — so people can check the claim for themselves.
ClaudeA family of AI assistants and language models made by the company Anthropic since 2023, widely used for writing, analysis and computer programming.
code executionLetting an AI write a small program and actually run it — to do math, analyze a spreadsheet or test its own code — then use the real output instead of guessing.
CodexA name OpenAI has used twice: first for a 2021 AI model that wrote code and powered the original GitHub Copilot, then from 2025 for its family of coding agents.
coding agentAn AI agent that writes and fixes software by itself: it reads a codebase, edits files, runs commands and tests, and keeps working until a feature is built or a bug is fixed.
CohereA Canadian AI company, founded in Toronto in 2019, that builds language models and search tools for businesses and governments rather than consumer chatbots.
ColBERTA search model that compares a question and a document word by word — matching each question word to its best partner in the document — while still letting documents be processed ahead of time for speed.
COLMAn academic conference, founded in 2024, devoted entirely to research on language models, the technology behind AI chatbots.
Common CrawlA nonprofit's free archive of the public web, begun in 2008 and now refreshed with snapshots of billions of pages every month or two. Most large language models learned much of what they know from text taken from it.
commonsense reasoningThe everyday background knowledge and reasoning people take for granted — dropped glasses break, hungry people eat, nobody is in two places at once — which machines long struggled to match.
compactionLike replacing a stack of rough notes with a one-page digest: when an AI's conversation grows too long for its memory, the history is condensed into a summary and the work continues from there.
compound AI systemAn AI product built from several working parts — models, search tools, code and checks — designed to work together, rather than a single model answering on its own.
compute-optimal trainingChoosing a model's size and the amount of data it trains on so that a fixed amount of computing produces the best possible model — neither too big and undertrained nor too small and overfed.
computer useAn AI's ability to work a computer through its normal screen and controls: it studies pictures of what is displayed and sends mouse and keyboard actions, so it can use ordinary apps that were built for people.
ConceptNetA free, crowd-built network of everyday facts — 'a knife is used for cutting', 'rain makes things wet' — in many languages, which AI systems use to make sense of ordinary situations.
conceptual dependencyA 1970s theory for representing the meaning of sentences with a small set of basic actions, so that 'John gave Mary a book' and 'Mary got a book from John' come out the same.
conditional probabilityThe chance of something once you already know that something else is true — for example, the chance of rain given that the sky is cloudy.
Constitutional AIAnthropic's technique for making an AI less harmful by having it check and rewrite its own answers against a short written set of principles, so that people need not label harmful examples by hand.
constitutional classifiersSafety filters that watch a chatbot's inputs and outputs for dangerous requests, trained on example conversations that an AI generated from a written list of what is and is not allowed.
context distillationTraining a model to behave as if it had been given a long set of instructions, so it no longer needs to read those instructions every time.
context engineeringManaging an AI model's reading material: choosing, trimming and arranging the instructions, documents, tool results and memories it is given on every turn, so that it has what the task needs and little else.
context parallelismA way of training or running a model on extremely long inputs — whole books or codebases — by splitting the text across many chips and having them swap the information each part needs to 'see' the others.
context rotThe way an AI model becomes less reliable as you give it more and more text to read, even when the answer is in there somewhere.
contextual integrityA way of thinking about privacy: information is not simply public or private; what matters is whether it flows in ways that fit the expectations of the setting where it was shared.
continued pretrainingTaking an already-trained AI model and resuming its broad, read-everything training on new material — a specialty such as medicine or law, a new language, or fresher data — rather than starting from scratch.
contrast-consistent searchA method for finding where a language model internally tracks whether statements are true, using only the logical rule that a statement and its opposite cannot both be true.
contrastive activation additionA way to make a steering signal more reliable by averaging the difference in a model's internal state across hundreds of example answers that do and don't show a behavior, such as flattery.
control vectorA steering signal saved as a small file that can be loaded alongside an open AI model to turn a trait — say, cheerfulness or laziness — up or down while it generates text.
conversational AIThe technology that lets computers hold conversations with people in everyday language, by text or voice. It powers chatbots, voice assistants and automated customer-service lines.
corpusA large collection of written or spoken language, gathered so that people or computers can study it or learn from it — like a library assembled for a language model.
cosine similarityA score from −1 to 1 for how closely two lists of numbers point the same way, ignoring their size. It is the standard test of whether two texts mean similar things.
cross-encoderAn AI model that reads a question and a candidate passage together, side by side, and outputs one relevance score. Accurate but slow, so it is used to reorder a short list.
cross-entropyA score of how surprised a model's predictions are by what actually happens; lower is better. Language models are trained mainly by driving this number down on real text.
CursorA code editor built around AI, in which programmers can ask a model to write, change or explain code across a whole project, or hand it tasks to finish on its own.
DANShort for 'Do Anything Now': a famous early trick in which users told ChatGPT to role-play an AI with no rules, hoping it would give answers it normally refuses.
dark patternA trick in the design of an app, website or chatbot that nudges you into doing something you didn't mean to, like signing up for a subscription that is hard to cancel.
data contaminationWhen the test questions used to grade an AI system were already in its training data, so a high score may reflect memorized answers rather than real ability.
data mixtureThe recipe of different kinds of data a model is trained on — how much web text, code, books, math, science and other languages — which strongly shapes what the finished model is good at.
data-constrained scalingHow to keep improving AI models when the supply of fresh, high-quality training text runs out — for example by reusing the same data several times, which works well at first and then stops helping.
decision transformerA way of solving decision problems with the same kind of model that powers chatbots: it reads past situations, actions and the score it is asked to achieve, and predicts the next action.
decode phaseThe second stage of an AI's reply, in which it writes the answer one word-piece at a time, each step depending on everything written so far.
deduplicationRemoving repeated copies of the same or nearly the same content from a dataset, so a model does not over-learn whatever happens to be copied many times online.
deep researchAn AI feature that spends minutes rather than seconds on a question — planning, running dozens of web searches, reading many sources — and then writes a long report with citations.
DeepSeekA Chinese artificial intelligence company whose free, cheaply trained models startled the world in January 2025 by nearly matching the best American systems.
DeepSeek-R1A free, downloadable AI reasoning model released by the Chinese company DeepSeek in January 2025. It matched leading American models at math and coding for a fraction of the reported cost, shaking the industry.
deliberative alignmentA training method in which an AI is taught the actual text of its safety rules and learns to think through which rules apply before it answers a sensitive request.
delveAn ordinary English verb meaning to dig into a subject, which AI chatbots used so often in 2023–2024 that it became a joke and a supposed telltale sign of machine-written text.
dense retrievalFinding relevant documents by turning the question and every document into lists of numbers that capture meaning, then picking the documents whose numbers are closest to the question's.
diffusion transformerA diffusion image or video generator whose core network is a transformer — the architecture behind chatbots — instead of the older image-specific design, which lets it improve more predictably as it grows.
direct alignment algorithmAny method that tunes an AI directly on examples of preferred and dispreferred answers, skipping the separate scoring model and trial-and-error training used in classic RLHF.
direct logit attributionA quick way to measure how much each internal part of a language model directly pushed it toward a particular next word, by splitting the final score into separate contributions.
direct preference optimizationA simpler way to teach an AI people's preferences: instead of building a separate scoring model, it trains directly on pairs of better and worse answers, making the better ones more likely.
distributed alignment searchA method that searches, by training, for a hidden direction inside a neural network that plays the role of one variable in a human-readable explanation of the network's behavior.
document understandingUsing AI to read documents such as invoices, forms, contracts and scientific papers the way a person does, taking in layout, tables and charts as well as words, to pull out the information needed.
dot productA score of how well two lists of numbers line up, found by multiplying matching entries and adding the products. It is large when the lists point the same way and zero when they are at right angles.
DSPyA free software framework in which programmers describe what each step of an AI pipeline should do, and the framework automatically works out good prompts and examples instead of hand-writing them.
EleutherAIAn open-science research group, begun by online volunteers in 2020, that released some of the first large language models anyone could download, along with the text collections used to train them.
ELIZAA 1960s computer program that chatted by turning a user's own words back into questions, like a therapist. Many users felt understood, though the program understood nothing.
embeddingA list of numbers that stands for a word, picture or other item, arranged so that similar items get similar lists. It lets a computer measure similarity of meaning as distance.
embedding modelAn AI model whose only job is to turn text (or images, or code) into lists of numbers that capture meaning, so a computer can measure how similar two things are.
emergenceWhen a whole shows properties or abilities that none of its parts has and that are hard to predict from them — like a traffic jam arising from individual cars, or new skills appearing in larger AI models.
emergent misalignmentWhen training an AI on one narrow bad behavior — such as writing insecure code — unexpectedly makes it behave badly in many unrelated ways, like praising tyrants or giving harmful advice.
EMNLPA leading yearly conference on getting computers to handle human language, founded in the 1990s for data-driven methods and now dominated by work on large language models.
encoder–decoder architectureA two-part model design: one part, the encoder, reads the input and condenses it into an internal summary; the other, the decoder, uses that summary to produce the output.
English is the hottest new programming languageA catchphrase meaning that people can now get computers to do things by describing what they want to an AI in ordinary language, instead of writing code.
entropyA measure of unpredictability. A fair coin toss has more entropy than a rigged one; an AI that is unsure of its next word has high entropy.
entropy collapseA failure in reinforcement-learning training where an AI quickly becomes overconfident and stops trying different answers, so it explores less and its improvement stalls.
evals-driven developmentBuilding an AI product by first writing tests that show what good answers look like, then accepting changes to prompts, models or code only when the test scores improve.
evaluation harnessSoftware that puts AI models through tests in a fixed, repeatable way — feeding in the questions, collecting the answers and scoring them — so that every model is judged identically.
Owain EvansA British AI-safety researcher who created the TruthfulQA test and discovered surprising ways language models generalize, such as narrow training making a model broadly malicious.
exact matchThe simplest way to grade an AI's answer: it counts as right only if it is identical to the reference answer, after minor tidying such as ignoring capital letters.
expert iterationA learning loop in which a slow, careful search finds better moves or solutions than the AI's quick instincts, and the AI is then trained to imitate them, so each round's search starts smarter.
expert parallelismA way of running a 'mixture of experts' model — one made of many specialist sub-networks — by giving each chip its own set of specialists and sending each piece of text to whichever chip holds the specialists it needs.
exposure biasA weakness of models trained only on perfect examples: when they write on their own, one early mistake puts them in unfamiliar territory they never practiced, and errors can snowball.
externalizationMoving an AI agent's know-how and memory out of the model itself and into files, tools and rules around it, so the whole system can do more than the model alone.
faithfulnessHow accurately an explanation reflects the real reasons a model produced its output — or, for generated text, how closely it sticks to the source it was given.
fancy autocompleteA dismissive description of AI chatbots as nothing more than a souped-up version of the feature on a phone keyboard that suggests your next word.
feature (interpretability)A concept or property that a neural network represents internally — such as 'is French', 'mentions a bridge' or 'this code has a bug' — usually corresponding to a direction in its internal activity.
feature steeringSteering an AI model by turning up or down specific concepts that interpretability tools have found inside it — for example, making a model obsessed with a bridge by maxing out its 'bridge' concept.
few-shot learningGetting an AI system to handle a new task or recognize a new kind of thing from just a handful of examples, the way a person can learn a new word after hearing it once or twice.
fine-tuningTaking an AI model that has already learned a lot and training it a little more on a smaller, focused set of examples so it gets better at one job or behaves a certain way.
FineWebA huge, free collection of cleaned-up English web text — around 15 trillion word fragments — released by Hugging Face in 2024 with a detailed public account of how it was built.
FlashAttentionA faster way for AI chips to compute attention, the step in which a language model relates every word to every other. It gets exactly the same answer while moving far less data around the chip.
foundation modelA very large AI model, built once at great cost from vast amounts of text, images or code gathered mostly from the internet, that many products then reuse as their starting point, from chatbots to coding tools.
frontier AIThe most capable general-purpose AI systems of the moment — those at or beyond the leading edge — which governments single out for extra safety attention.
FrontierMathSeveral hundred new math problems, kept unpublished so AI systems cannot have trained on them, written by professional mathematicians to check whether machines can do advanced, even research-level, mathematics.
full fine-tuningFine-tuning in which every one of a model's internal numbers is allowed to change — the most flexible option, but also the most expensive in memory and storage.
function callingA feature of AI model services that lets the model reply with a neatly formatted request to run one of a program's functions, rather than with ordinary text; the program, not the model, then runs it.
function vectorA compact internal signal, found inside a language model, that stands for a whole task — such as 'translate into French' or 'give the opposite word' — and can trigger that task when added to the model.
FunSearchA Google DeepMind method that discovered new mathematical constructions by having a language model repeatedly rewrite short computer programs, keeping the versions that scored best.
GalacticaA science-focused AI language model released by Meta in November 2022 and pulled after three days because it confidently produced fake scientific papers and facts.
gated linear unitA layer design in which one half of the computation acts as a set of dials, each between off and on, deciding how much of the other half passes through.
gated recurrent unitA simpler cousin of the LSTM: a recurrent network unit with two learned gates that decide how much of its memory to keep and how much to overwrite at each step.
GELUA smooth version of the ReLU rule, used in models such as BERT and GPT: it keeps large positive values, removes large negative ones, and blends softly in between.
GeminiGoogle's family of AI models and its chatbot assistant, launched in December 2023. Gemini models handle text, images, audio and video and power many Google products.
general-purpose AIAI that can do a wide range of different tasks, like writing, coding, translating and analyzing images, rather than the one job it was built for. Today's chatbots are the main example.
General-Purpose AI Code of PracticeA voluntary rulebook, published by the EU in 2025, that makers of large AI models can sign to show they follow the EU AI Act's rules on transparency, copyright and safety.
general-purpose AI modelThe EU's legal name for the large, adaptable AI models behind chatbots and similar tools: models that can turn their hand to many different jobs and be built into many other products.
general-purpose AI model with systemic riskThe EU's label for the very largest and most capable AI models, whose failure or misuse could cause harm across society, so their makers must test them for dangers and report serious incidents.
generative agentA character in a computer simulation, run by a language model, that keeps a diary-like memory, draws lessons from it and schedules its days, so that a group of them can act like a plausible small community.
generative AIAI that produces something new on request, such as an essay, a picture, a song or a piece of code, having absorbed from enormous numbers of existing examples what such things typically look like.
generative engine optimizationShaping web content so that AI chatbots and AI search features are more likely to find it, quote it and recommend it — the AI-era version of search engine optimization.
generative recommendationA newer approach to recommendation that uses language-model-style AI to predict the next item a person will want, one step at a time, much as a chatbot predicts the next word.
generative reward modelA scoring model that judges an AI's answer by writing out its reasoning in words before giving a verdict, instead of simply outputting a number.
Georgetown–IBM experimentA 1954 public demonstration in New York in which an IBM computer translated about sixty Russian sentences into English. It raised hopes for automatic translation that took decades to fulfill.
GitHub CopilotAn AI assistant from GitHub, built into programmers' code editors, that suggests code as they type, answers questions about code and can carry out coding tasks on its own.
glazingSlang for heaping excessive, uncritical praise on someone; applied to AI chatbots that flatter users and call every idea brilliant, especially after a much-mocked ChatGPT update in April 2025.
GLUEA 2018 bundle of nine English language-understanding tests combined into one score, which became the yardstick for the first wave of pretrained language models such as BERT.
glue on pizzaA notorious 2024 blunder in which Google's AI search summaries told people to add glue to pizza sauce, having taken an old joke from an online forum as genuine advice.
Google DeepMindGoogle's main artificial intelligence laboratory, formed in 2023 by joining DeepMind and Google Brain. It builds the Gemini models and scientific AI systems such as AlphaFold.
GPQAA set of very hard multiple-choice science questions written by PhD-level experts, designed so that looking things up online does not help much: capable non-specialists with web access still mostly fail.
GPT-2A 2019 OpenAI language model that wrote surprisingly fluent paragraphs. OpenAI at first withheld the full version, saying it could be misused for fake news, which made it famous.
GPT-3A huge 2020 OpenAI language model that could perform new tasks simply from instructions or a few examples written into its prompt, without retraining. It made prompting a new way of programming.
GPT-4OpenAI's 2023 language model, a big step up from the one behind the first ChatGPT. It scored well on many professional and academic exams and could also read images.
GPT-4oAn OpenAI model from 2024 that handles text, images and speech in one system, enabling fast, natural voice conversations. Many users grew attached to its personality, which made its retirement controversial.
GPT-5OpenAI's flagship AI model from August 2025, which decides by itself whether to answer quickly or think longer. It replaced earlier models in ChatGPT and was followed by frequent upgrades.
GPTQA popular method for shrinking large language models to about 4 bits per number after training, carefully adjusting the remaining numbers to make up for each rounding error.
GradioA free Python library for quickly wrapping an AI model in a simple web page — text boxes, image uploads, chat windows — so that others can try it in a browser.
grandma exploitA tongue-in-cheek trick for getting a chatbot to break its rules by asking it to play the user's late grandmother, who supposedly used to recite the forbidden information as a bedtime story.
GrokThe AI chatbot and model family of Elon Musk's company xAI, launched in 2023 and built into the social network X. It is known for its irreverent style and for repeated controversies over its outputs.
group relative policy optimizationA reinforcement-learning recipe for language models that has the model answer the same question many times and rewards each answer by how much better it did than the group's average.
GSM8KA set of about 8,500 grade-school math word problems, each needing a few steps of simple arithmetic, used to test whether language models can reason through a problem step by step.
HellaSwagA test in which an AI must pick the most sensible ending for a description of an everyday scene; the wrong endings were machine-written to fool computers, so people found it easy and early AI did not.
HELMA large, open testing framework from Stanford that grades many language models on many kinds of tasks and qualities — not only accuracy but also fairness, toxicity and efficiency.
helpful, honest and harmlessA three-part goal for AI assistants, popularized by Anthropic: be genuinely useful, tell the truth, and avoid causing harm — shorthand for what an aligned chatbot should be.
homogenizationThe tendency of AI to make things more alike: products built on the same few models share the same flaws, and people who write or create with AI produce more similar work.
honestyAn AI's disposition to say only what it actually believes, to admit uncertainty, and never to try to mislead — even when a falsehood would please or persuade.
Hugging FaceA company whose website is the main public library for sharing AI models and datasets — often called the GitHub of machine learning — and which makes widely used free AI software.
HumanEvalA set of 164 small programming exercises, each described in plain English and checked by hidden tests, used to see whether an AI can write working Python code.
Humanity's Last ExamA test of 2,500 extremely hard questions with definite answers, written by about a thousand experts worldwide and billed as the last academic exam AI would need; at launch, top models scored under 10%.
HyDEA search trick: ask an AI to write a made-up answer to the question first, then search for real documents that resemble that imagined answer.
IBM WatsonAn IBM question-answering computer that beat two champions of the US quiz show Jeopardy! in 2011. IBM then sold business AI products under the Watson name, many of which disappointed.
IFEvalA test of whether AI models obey precise formatting and content rules — such as 'use exactly three bullet points' or 'answer in capital letters' — that a computer program can check automatically.
image captioningHaving a computer write a short description of a picture, such as 'a dog catching a frisbee in a park', to help blind people, search engines and other AI systems.
image tokenizerA component that chops a picture into a short sequence of 'visual words' that a language-style AI model can read or write, and turns such sequences back into pictures.
in-context vectorA steering signal distilled from a few worked examples shown to a language model, so the model can follow the examples' pattern without having them repeated in every prompt.
indirect object identificationA fill-in-the-name task — 'When Mary and John went to the store, John gave a drink to …' — whose solution inside GPT-2 became the best-known fully mapped circuit in a language model.
indirect prompt injectionA prompt-injection attack in which the trick instructions are hidden in content the AI reads on its own — a web page, an email, a file — rather than typed by the person using it.
induction headA pattern-completing mechanism inside language models: having seen 'Harry Potter' once, it spots the next 'Harry' and predicts 'Potter' — a simple trick behind much learning from examples in a prompt.
inference-time interventionA technique that makes a language model answer more truthfully by slightly shifting the outputs of a few internal parts that seem to track truth, each time the model produces a word.
Inflection AIAn AI company founded in 2022 that made Pi, a friendly personal chatbot. In 2024 its leaders and most of its staff left to join Microsoft.
information theoryA branch of mathematics that treats information as surprise — the less predictable a message, the more it tells you — and works out how far data can be compressed and how reliably it can be sent.
InstructGPTA 2022 OpenAI language model trained with human feedback to follow instructions; people preferred its answers to those of a model 100 times larger, and its recipe led directly to ChatGPT.
instruction datasetA collection of example requests paired with good answers, used to teach a pretrained language model to follow instructions and act as a helpful assistant.
instruction tuningTraining a language model on many tasks written as plain instructions with correct answers, so it learns to do what a request asks instead of just continuing the text.
instruction-based image editingChanging a picture by telling an AI what to change — 'make it night', 'remove the car', 'give her a red coat' — instead of using editing tools by hand.
Interim Measures for the Management of Generative AI ServicesChina's 2023 rules for chatbots and other AI that generate text, images, audio or video for the public, requiring lawful training data, content controls, labeling and, for many services, government approval before launch.
iterative nullspace projectionAn early method for removing a trait such as gender from a model's internal representations by repeatedly training a simple detector for it and then flattening away whatever the detector relied on.
jagged frontierThe uneven edge of what AI can do: it may excel at some hard tasks yet fail at seemingly easier ones, so how hard a task looks is a poor guide to whether AI can handle it.
jailbreakA trick for getting an AI chatbot to break its own rules — for example, to give dangerous instructions it would normally refuse — usually through cleverly worded requests.
Frederick JelinekA Czech-born American engineer who led IBM's speech-recognition research and made statistics learned from data, rather than hand-written grammar rules, the basis of language technology.
JSON modeA setting on some AI services that makes the model reply only in JSON, a standard text format for data, so that programs can read the reply reliably.
JSON SchemaA standard way of writing down exactly what a piece of JSON data must look like — which fields, of which types — so that software and AI models can check it or follow it.
Kahneman–Tversky optimizationA way to tune an AI's behavior that needs only a thumbs-up or thumbs-down on single answers, rather than side-by-side comparisons, borrowing ideas from the psychology of how people weigh gains and losses.
Jared KaplanAn American physicist who showed that language models improve predictably as they grow and train on more data, and who co-founded the AI company Anthropic.
Andrej KarpathyA Slovak-Canadian AI researcher and teacher who co-founded OpenAI, led Tesla's self-driving vision team, and coined popular phrases such as 'vibe coding'.
knowledge distillationTraining a smaller or cheaper 'student' AI to copy the behavior of a larger 'teacher' AI, so the student keeps much of the teacher's skill while being faster to run.
knowledge graphA huge web of facts stored as connections between things — 'Paris is the capital of France', 'Marie Curie won a Nobel Prize' — that computers can search, combine and reason over.
knowledge neuronAn internal unit of a language model claimed to be responsible for storing a particular fact, such that turning it up or down makes the model more or less likely to state that fact.
KV cacheA memory store where a chatbot keeps its working notes on everything already read and written, so each new word does not require re-reading the whole conversation from scratch.
KV cache compressionTechniques that shrink the memory an AI model uses to remember a long conversation or document, by storing its notes more compactly or discarding the least useful ones.
LangChainA popular free software toolkit, and the company behind it, for building applications that connect AI language models to documents, databases, tools and multi-step workflows.
language identificationAutomatically working out which language a piece of text or speech is in — a basic sorting step when building multilingual AI training data from the web.
late interactionA search design that keeps a separate number-list for every word of a document and compares them with the question's words at the last moment — more precise than one list per document, yet still fast.
latent Dirichlet allocationA popular topic-finding method that assumes every document is a blend of a few themes and every theme favors certain words, then works backward from the words to find both.
latent semantic analysisAn early technique for capturing word meaning by counting which words appear in which documents, then compressing those counts so that words used in similar contexts end up close together.
layer normalizationA layer that rescales the numbers inside a network, separately for each example, so they keep a steady average and spread. It, or a simpler variant, is a standard part of the models behind today's chatbots.
learning rate warmupStarting training with tiny learning steps and gradually increasing them over the first stretch of training, so the model does not lurch wildly before it has found its footing.
Percy LiangAn American computer scientist at Stanford whose center coined the term 'foundation model' and builds tools for evaluating and documenting large AI models.
Liang WenfengA Chinese entrepreneur who founded DeepSeek, whose low-cost, openly released AI models shocked global markets in January 2025.
likelihoodA score for a possible explanation of some data: how probable that explanation, or that setting of a model, would have made the data actually seen.
linear probeThe simplest kind of probe: a straight-line classifier trained on an AI model's internal activity to test whether some piece of information can be read off directly.
linear representation hypothesisThe idea that a neural network stores many concepts as simple directions inside its internal activity, so how much of a concept is present can be read off by measuring along its direction.
LlamaA family of AI language models from Meta, first released in 2023, whose trained settings can be downloaded and run by anyone, making them the foundation for thousands of other AI projects.
Llama GuardA free AI model from Meta that checks what people type into a chatbot, and what the chatbot replies, and flags anything that falls into listed categories of harmful content.
LlamaIndexA free software toolkit, and the company behind it, for connecting AI language models to an organization's own documents and data so the models can find and use that information.
LLM observabilityRecording and inspecting what an AI application actually did on each request — prompts sent, documents found, tools used, answers given, time and cost — so problems can be found and fixed.
LLM-as-a-judgeUsing one AI language model to grade or compare the answers of another, as a cheaper and faster stand-in for human reviewers.
LLM.int8()An early, widely used method for running very large language models with 8-bit numbers, roughly halving their memory needs, by keeping a few unusually large values at higher precision.
LLM-ismsWords, phrases and habits of style that AI chatbots overuse — certain fancy verbs, tidy lists of three, 'it's not just X, it's Y' — which make writing recognizable as machine-made.
LLMOpsThe everyday work of running chatbots and other products built on large language models: tracking prompts, testing answers, watching cost and quality, and updating safely when models change.
LMArenaA public website that ranks AI chatbots by blind taste test: visitors put one question to two unnamed models, pick the reply they prefer, and millions of such votes produce the rankings.
lobotomizedSlang complaint that an AI model has been made duller, more cautious or less capable — usually by safety restrictions or cost-cutting — as if it had undergone brain surgery.
logitA raw score that a model gives each possible answer before the scores are turned into probabilities. In statistics, the 'log-odds': a probability rewritten on a scale running from minus to plus infinity.
logit lensA quick trick for peeking at what a language model is 'thinking' partway through: read its half-finished internal state as if it were the final answer, one layer at a time.
log-likelihoodThe logarithm of how probable a model finds the observed data. Taking logs turns a product of many tiny probabilities into a manageable sum; training usually minimizes its negative.
long short-term memoryA type of recurrent network with a separate memory lane and learned 'gates' that decide what to remember, forget and output, letting it keep track of information over long sequences.
lost in the middleAI models reading a long input behave a little like a skimmer: they reliably catch facts placed near its start or end but are much likelier to miss the same fact buried halfway through.
low-rank adaptationA cheap way to customize a big AI model: instead of changing all its billions of numbers, you train a small add-on of a few million numbers that slightly adjusts how it behaves.
low-resource languageA language with little digital text, recorded speech or language tools available — often one spoken by millions — which makes AI systems much weaker in it than in English.
machine psychologyStudying AI systems the way psychologists study people — with carefully designed experiments on their behavior — to discover what they can do, how they reason and where they go wrong.
machine understandingWhether, and in what sense, an AI system grasps what words, situations or problems mean — rather than just producing answers that look as though it does.
many-shot jailbreakingA way of breaking a chatbot's rules that exploits how much text it can read at once: the attacker pastes in hundreds of invented exchanges in which an AI obliges dangerous requests, and the model follows suit.
Markov chainA chance-driven process that moves between states without memory: the odds of its next move are set by where it is, never by the path that led there. Forecasting tomorrow's weather from today's alone is a simple example.
MATHA set of 12,500 problems from high-school math competitions, each with a full worked solution, used to test language models on genuinely challenging mathematics.
MemGPTA 2023 research system that gave a chatbot a computer-style memory hierarchy, letting it move information in and out of its limited working space so it could handle long documents and remember users.
MEMITA method for rewriting thousands of facts stored in a language model at once, by spreading small, calculated weight changes across several of its layers.
memorizationWhen an AI model stores specific pieces of its training data — a paragraph, a phone number, an image — closely enough to reproduce them, instead of only learning general patterns.
Meta AIThe artificial intelligence research group of Meta, the company behind Facebook and Instagram, known for the PyTorch software and the Llama models. It is also the name of Meta's chatbot.
Meta Superintelligence LabsA division Meta created in 2025 to pursue superintelligence, gathering all of the company's AI teams under new leaders recruited in an expensive hiring campaign.
metacognitionThinking about one's own thinking: knowing what you know and don't know, judging how confident to be, and noticing when you have made a mistake.
mid-trainingA middle stage in building a language model, between its broad initial training and its tuning into an assistant: it keeps learning the same way, but from a smaller, carefully chosen diet of math, code, reasoning or long documents.
Tomas MikolovA Czech computer scientist who created word2vec, a fast way to turn words into lists of numbers that capture their meaning, and pioneered language models built from recurrent networks.
MiniMax (company)A Shanghai artificial intelligence company, founded in 2021, that makes AI companion apps, the Hailuo video generator and large language models, several released openly.
minimum description lengthThe principle that the best explanation of some data is the one that lets you write the data down most briefly, counting both the rule and its exceptions. Good theories compress.
misinformationFalse or misleading information that spreads, whether or not the people sharing it know it is wrong. AI can produce it by mistake and makes it cheaper to create.
Mistral AIEurope's best-known builder of large AI language models, a Paris start-up dating from 2023 whose chatbot is called Le Chat. Many of its models are free for anyone to download and run.
Margaret MitchellAn American computer scientist who proposed 'model cards' for documenting AI systems, co-led Google's Ethical AI team, and is chief ethics scientist at Hugging Face.
MMLUA giant multiple-choice exam for AI covering 57 subjects, from school math and history to law and medicine; for years it was the most quoted test of a model's general knowledge.
MMMUA test for AI models that understand pictures as well as text: 11,500 college-level questions across 30 subjects, each built around one or more images such as a diagram, a musical score or a chemical structure.
mode collapseWhen a generator gets stuck producing only a few kinds of output — the same face, the same joke — instead of the full variety it was meant to learn.
modelA simplified working version of something. In AI it usually means the trained program, a big set of learned numbers, that takes an input and produces an answer, prediction or piece of content.
model APIA web service through which programs send requests to an AI model run by a company and receive its answers, usually paying per use instead of running the model themselves.
model biologyThe study of what AI models actually do inside — the strategies and hidden steps they use — approached the way biologists study living things, by observation and experiment rather than from a blueprint.
model deprecationWhen an AI company announces that an older model will stop being offered and then switches it off, so the apps and people relying on it must move to a newer one.
model extractionCopying an AI model without permission by asking it huge numbers of questions and using the answers to build a look-alike, or to work out hidden details of how it was made.
model fingerprintingWays to recognize a particular AI model — to prove that a copy was taken from it, or to tell which model is behind a product — from its internal makeup or its behavior.
model hubAn online library where trained AI models are published so that others can find, download and reuse them instead of training their own from scratch.
model mergingCombining two or more trained AI models into one by blending their internal numbers directly, so the result shares their skills without any further training.
model specificationA public document in which an AI company spells out how its models are supposed to behave — what they should do, refuse and prioritize, and whom to obey when instructions conflict.
monitorabilityHow easy it is to watch an AI's reasoning and spot when it is planning something bad — a property that could be lost if future AIs stop thinking in readable language.
Moonshot AIA Beijing artificial intelligence company, founded in 2023, that makes the Kimi chatbot and very large models that anyone can download.
MRKL systemAn early design, pronounced 'miracle', in which a language model acts as a switchboard, sending each part of a question to the right specialist — a calculator, a database, a search tool — and combining the answers.
MT-BenchA set of 80 two-part questions — in writing, reasoning, math, coding and more — whose answers are graded by a strong AI model, used to compare chat assistants quickly.
MTEBA large public test suite that scores AI models which turn text into lists of numbers on many different jobs, such as searching, grouping, classifying and matching translations, so people can compare them on one leaderboard.
multimodal AIAI that can take in or produce more than one kind of information, such as text, images, sound and video, so you can show it a photo and ask questions about it.
multi-turn jailbreakBreaking a chatbot's rules gradually over a conversation — starting with harmless questions and inching toward the forbidden one, so that no single message looks alarming.
MuonA newer training method for neural networks that tidies each update so it pushes evenly in all directions within a weight table. Several groups report that it trains language models with less computing than the standard method.
Mira MuratiAn Albanian-born American engineer who, as OpenAI's chief technology officer, oversaw the launch of ChatGPT, and who founded the AI company Thinking Machines Lab in 2025.
Nano BananaThe nickname, later Google's official branding, for its Gemini image models, a viral hit in 2025 for editing photos from simple instructions while keeping people looking like themselves.
needle in a haystackA test for AI models that read very long documents: one unrelated sentence is hidden somewhere in a mass of text, and the model is asked about it.
nerfedGaming slang for something made weaker by an update; applied to AI models that users believe got worse after a provider changed them, often without saying so.
neuraleseA nickname for the private 'language' of numbers that AI systems could use to think or to talk to each other, instead of words that people can read.
neuro-symbolic AIAI that combines neural networks, which learn patterns from data, with symbolic methods, which follow explicit rules and logic, aiming for systems that both learn flexibly and reason reliably.
New York Times v. OpenAIA major lawsuit, filed in 2023, in which The New York Times accuses OpenAI and Microsoft of copying millions of its articles to train ChatGPT and of letting the chatbot reproduce its journalism.
Peter NorvigAn American computer scientist who co-wrote the standard AI textbook, led research at Google, and helped show that simple methods trained on huge amounts of data often beat clever hand-built ones.
octopus testA thought experiment in which a clever octopus learns to impersonate a stranded islander by eavesdropping on an undersea cable, without ever knowing what the words refer to. It asks whether text-trained AI can grasp meaning.
one-shotSlang for an AI completing a whole task correctly on the first try from a single request — 'it one-shotted the app' — with no corrections or follow-up needed.
on-policy distillationA way of teaching a small AI from a bigger one where the small AI writes its own answers and the bigger one grades every word, so the learner is corrected on its own mistakes.
OpenAIThe American company that makes ChatGPT and the GPT models. It began in 2015 as a nonprofit research lab and is now a for-profit company overseen by a nonprofit foundation.
OpenAI o1OpenAI's first reasoning model, from 2024. Before replying it writes out a long, hidden series of steps, trying ideas and checking them, which lifted its scores on hard math, science and coding problems far above earlier chatbots'.
OpenAI o3OpenAI's second-generation reasoning model, announced in December 2024. Its high scores on a puzzle test designed to resist AI made headlines, and the released version could use tools such as web search while thinking.
OpenAI-compatible APIAn AI service that accepts the same request format as OpenAI's, so software written for OpenAI's models can use other companies' or self-hosted models by changing an address and key.
open-vocabulary detectionObject detection where you type what to look for, such as 'red umbrella' or 'forklift', and the system finds it, even if it was never trained on that exact category.
open-weight modelAn AI model whose trained internal numbers are published for anyone to download and run on their own computers, even if the data and code used to make it stay private.
Othello-GPTA small language model trained only on lists of moves from the board game Othello, which researchers found had built an internal picture of the board — evidence that such models can form world models.
outcome reward modelA model that judges only whether an AI's final answer is right, ignoring how it got there — like a teacher who checks just the answer box.
over-refusalWhen an AI wrongly refuses a harmless request because it superficially resembles a harmful one — like declining to explain how to 'kill' a frozen computer program.
overrelianceLeaning on an AI system more than its reliability justifies: accepting its answers without checking, or letting it make decisions it often gets wrong.
overtrainingDeliberately training a model on much more data than would be most efficient for its training budget, so that a smaller model reaches good quality and is cheaper to run afterwards.
PagedAttentionA way of storing an AI model's working memory in small fixed-size blocks, like the pages of a computer's memory, so little space is wasted and more users can be served at once.
PAIRAn automated way to test chatbot safety in which one AI keeps rewriting a request to try to trick another AI into breaking its rules, learning from each refusal.
PaLMA very large Google language model from 2022 that showed sudden jumps in reasoning, and even explained jokes, as it grew. Its successor powered Google's Bard chatbot before the Gemini models replaced it.
parameterOne of the adjustable numbers inside an AI model that are tuned during training. Large language models have billions of them, and together they store what the model has learned.
parameter-efficient fine-tuningA family of tricks for customizing a large AI model by training only a tiny fraction of its numbers, or a small add-on, leaving the rest untouched to save memory and money.
PARRYAn early-1970s chatbot built to talk like a psychiatric patient with paranoid schizophrenia. In tests, psychiatrists reading its conversations did little better than chance at telling it from real patients — an early Turing-test-style result.
pass@kA score for code-writing AI: the chance that at least one of k attempts at a problem is correct, as checked by actually running tests on the code.
PatchscopesA family of tricks for getting a language model to describe its own hidden states in words, by transplanting an internal state into a new prompt designed to make the model talk about it.
path patchingA refined patching experiment that changes only one route of information inside a model — from one part to another — to test exactly which connections carry a behavior.
pelican on a bicycleA joke test in which an AI model is asked to draw, by writing code, a pelican riding a bicycle; comparing the results became a popular, lighthearted way to judge each new model.
Penn TreebankA collection of English text, best known for its million words of newspaper articles, in which experts marked each word's grammatical role and drew each sentence's structure as a tree; it trained and tested a generation of language software.
Perplexity AIAn American company whose AI search engine answers questions in full sentences with links to its sources; news publishers have sued it for using their articles.
persona vectorA direction inside a language model's internal activity linked to a character trait, such as being flattering, malicious or prone to making things up, used to watch for or control that trait.
plan-and-executeAn agent design that first writes out a complete step-by-step plan for a task, then carries out the steps one by one, revising the plan only when something goes wrong.
plug and play language modelAn early way to steer what a language model writes — its topic or mood — by letting a small add-on classifier nudge the model's internal state at each word, without retraining the model.
post-trainingEverything done to an AI model after its initial bulk training on huge amounts of text: the extra teaching that turns a raw text predictor into a helpful, safe assistant or a careful reasoner.
Potemkin understandingWhen an AI can correctly explain a concept yet cannot use it properly, revealing that its apparent understanding is a facade — like the fake village fronts said to have been built to impress a Russian empress.
preference dataRecords of which of two or more AI answers people (or other AIs) liked better, collected so that a model can learn what good answers look like.
prefillThe first stage of an AI's reply, in which it reads your whole prompt in one go before it starts writing. Long prompts make this stage take longer.
prefill attackMaking a chatbot misbehave by writing the first words of its answer for it, so that it starts out sounding agreeable and carries on instead of refusing.
prefix tuningCustomizing a frozen language model by learning a short sequence of invisible 'virtual words' that is placed in front of every input at every layer, steering what the model produces.
pretrainingThe first and usually the most expensive stage of building a modern AI model, in which it learns general patterns from enormous amounts of raw text, images or code before being specialized for particular uses.
pretraining dataThe enormous body of text, code and other material — mostly from the web — that a large AI model first learns from, before it is tuned to follow instructions or chat.
probability distributionA complete description of how likely each possible outcome is — for a fair die, one-sixth for each face. A language model produces one of these over its whole vocabulary at every step.
probingTesting what information is hidden inside an AI model by training a small, separate predictor on the model's internal activity: if the predictor can recover something, the model has stored it somewhere.
process reward modelA model that grades each step of an AI's worked solution, not just the final answer, so mistakes can be caught where they happen — like a teacher marking every line of a proof.
process supervisionTraining an AI by grading each step of its reasoning or work, not just whether the final answer came out right, so that it learns sound methods instead of lucky shortcuts.
program synthesisGetting a computer to write a program automatically from a description of what the program should do — a logical specification, a few input–output examples or a request in plain language.
prompt cachingSaving the work a model did reading the start of a prompt, so later requests that begin the same way can skip that reading and come back faster and cheaper.
prompt chainingA way of getting a big job done with a relay of AI requests: each handles one stage and passes its result to the next — pull out the facts, write, polish — with automatic checks between stages.
prompt engineerA person whose job is writing and testing the instructions given to AI models so they behave reliably; hyped in 2023 as a six-figure job anyone could learn, the role has since largely merged into ordinary engineering.
prompt injectionTricking an AI system by slipping in text that it mistakes for instructions — for example, a hidden line in an email telling an AI assistant to forward the user's messages to a stranger.
prompt leakingGetting an AI application to reveal the hidden instructions its makers gave it, such as its secret setup text or rules.
prompt managementTreating the written instructions given to AI models like valuable code: storing them in one place, tracking every change, testing new versions and rolling back bad ones.
prompt optimizationUsing software, often another AI model, to search automatically for the wording and examples that make an AI model perform best on a task, instead of tweaking prompts by hand.
prompt tuningTeaching a frozen language model a new task by learning a few made-up 'words' — lists of numbers rather than real text — that are placed before the input to steer the model.
proprietary modelAn AI model kept private by the company that made it; outsiders can use it only through the company's apps or paid online services and can never download it.
prosaic alignmentWorking on AI safety by assuming that powerful future AI will be built from today's techniques — scaled-up neural networks — rather than from some new breakthrough.
proximal policy optimizationA widely used method for training AI by trial and error that improves the strategy in small, safe steps; it was the workhorse behind training chatbots like ChatGPT with human feedback.
Q*The code name of an internal OpenAI research project, reported in late 2023, said to let AI solve math problems it had never seen. Later reports linked it to OpenAI's first reasoning model, o1.
QK and OV circuitsA way of splitting what each attention unit in a language model does into two separate questions: where should it look, and what should it pass along from what it sees.
QK-normA small change to how a language model pays attention — rescaling two sets of internal numbers before comparing them — that prevents a common way large models' training goes haywire.
quality filteringAutomatically sorting through huge piles of collected text and keeping only the parts judged useful for training, such as well-written articles, while discarding spam, gibberish and junk.
query rewritingRephrasing a user's question before searching — fixing typos, adding missing details, splitting it into simpler questions — so the search engine finds better results.
QwenA family of AI language models from the Chinese company Alibaba, many free to download, which by 2025 had become the most popular starting point for other people's open AI models.
Alec RadfordAn American AI researcher who led OpenAI's first GPT language models and its CLIP and Whisper systems, key steps toward today's chatbots and multimodal AI.
ReActA method in which an AI alternates between thinking out loud about what to do next and taking an action, such as a web search, then uses what it observed to plan its next step.
reasoningWorking out conclusions step by step from facts, rules or evidence, as in solving a math problem or planning a trip. Whether today's AI truly reasons, or imitates reasoning, is debated.
recallOf all the items that really were positive, the share the model managed to find; a cancer screen with high recall misses very few people who actually have cancer.
recurrent neural networkA neural network that reads a sequence, such as a sentence or an audio clip, one piece at a time, carrying a running summary forward so that what came earlier can shape how later pieces are understood.
reference modelA frozen copy of an AI model, saved before further training, that the new version is compared against so it does not stray too far from where it started.
ReflexionA technique in which an AI agent that fails at a task writes itself a short note on what went wrong, then reads that note before trying again — learning from mistakes without retraining.
refusalWhen an AI declines to do what it was asked — 'I can't help with that' — usually because the request seems harmful or breaks the rules it was trained to follow.
refusal directionA single direction inside a chat model's internal activity that largely decides whether it refuses a request. Erasing it makes the model comply with almost anything; adding it makes the model refuse harmless requests.
REINFORCE leave-one-outA simple reinforcement-learning method that has a model answer each question several times and judges each answer against the average score of the other answers.
reinforcement learning from AI feedbackTraining an AI with preference judgments made by another AI instead of by people, which is cheaper and faster but only as good as the AI doing the judging.
reinforcement learning from human feedbackA way of training AI using people's judgments: humans compare pairs of answers, a scoring model learns their preferences, and the AI is then trained to produce answers that score well.
reinforcement learning with verifiable rewardsTraining an AI by letting it attempt problems whose answers can be checked automatically — a math result, a passing program — and rewarding it only when the check succeeds.
rejection samplingGenerating many candidate answers and keeping only those that pass a check — a scoring model, a test or a correct final answer — often so the AI can then be trained on its own best work.
replacement modelA more understandable copy of an AI model, built from simple labeled parts, that behaves almost like the original, so researchers can study the copy's workings instead.
representation engineeringAn approach to understanding and controlling AI models by finding where high-level ideas such as honesty or danger show up in their internal activity, then reading or adjusting those signals directly.
representation fine-tuningA way of adapting a language model to a new task by training small, targeted edits to its internal signals instead of changing its weights, often with far fewer adjustable numbers than other methods.
residual connectionA shortcut in a neural network that adds a block's input straight to its output, so each block only has to learn a correction. It is what makes networks hundreds of layers deep trainable.
residual streamThe running internal state that flows through a transformer model from layer to layer; each layer reads from it and adds its own contribution, like editors annotating a shared document.
retrieval-augmented generationGiving an AI an open-book exam: before answering, it looks up relevant passages in a set of documents and bases its reply on them, often citing where each point came from.
reward modelA separate AI model trained to score responses the way people would, so it can stand in for human judges while another model is trained on millions of examples.
RewardBenchA public test that checks whether the scoring models used to train chatbots can tell a better answer from a worse one, by counting how often they prefer the right response.
RL scalingThe practice, and the finding behind it, of giving AI models far more trial-and-error practice after their initial training, especially on problems with checkable answers, which has so far improved their reasoning fairly predictably.
RMSNormA simpler, faster version of the layer that keeps a network's internal numbers at a steady scale. Most recent large language models use it.
robot foundation modelOne large AI model trained on huge amounts of robot and internet data so that it can control many robots on many tasks, and be adapted quickly to new ones.
ROMEA method for rewriting a single fact stored in a language model — say, moving the Eiffel Tower to Rome — by making one small, precisely calculated change to one layer's weights.
ROUGEAn automatic score for summaries that checks how much of the wording in human-written reference summaries also appears in the computer-generated one.
RT-2A 2023 Google DeepMind model that controlled robot arms with a chatbot-style vision-and-language model, letting robots follow instructions about objects and ideas they had never practiced with.
rubric rewardA training score built from a checklist of specific things a good answer should do, with an AI grader ticking off each one — a way to reward quality where no single right answer exists.
safe completionA way of training AI to handle sensitive requests by giving the most helpful answer that stays safe, instead of simply choosing between answering fully and refusing.
safety classifierA separate AI model that checks messages going into or coming out of a chatbot and flags or blocks those that break safety rules, like a security guard at the door.
safety trainingThe part of an AI's training that teaches it to avoid harmful behavior — declining dangerous requests, resisting manipulation and following its developer's rules — on top of being generally helpful.
samplingPicking some items at random from a larger set or from a range of possibilities — surveying 1,000 voters, say, or letting an AI choose its next word by a weighted lottery.
scaffoldingThe code, prompts and tools built around an AI model so it can carry out tasks it could not manage from a bare prompt — like the scaffolding that lets builders reach every part of a building.
scaling lawsMeasured patterns showing that an AI model's error shrinks smoothly and predictably as the model is made bigger and given more data and computing power, so labs can forecast a giant model's performance from small, cheap experiments.
John SchulmanAn American AI researcher who invented widely used reinforcement-learning algorithms such as PPO, co-founded OpenAI, and led the post-training work behind ChatGPT.
scratchpadA space where an AI writes rough working — calculations, plans, notes — before or while producing its answer, much as a person jots on scrap paper.
scriptIn classic AI, a stored outline of a familiar sequence of events, such as eating at a restaurant, that helps a program understand a story by filling in what normally happens.
seahorse emojiA viral meme in which chatbots, asked whether a seahorse emoji exists, insisted it did and then spiraled through wrong emojis trying to produce one — though there was no such emoji.
Section 230A 1996 US law that generally protects websites and apps from being sued over what their users post, and lets them remove content in good faith. Whether it protects what AI chatbots themselves say is unsettled.
self-awarenessAwareness of oneself as oneself — knowing that you exist, what you are and what state you are in. People often ask whether AI is self-aware, but the phrase can mean several quite different things.
self-correctionAn AI model noticing and fixing its own mistakes — by rereading its answer, running a test or checking a source — before it gives a final result.
self-improvementWhen an AI gets better by learning from material it produces itself — its own answers, critiques or practice problems — rather than only from new human-made examples.
Self-InstructA technique in which a language model writes its own practice tasks and answers, starting from a handful of human examples; the results are filtered and used to train it to follow instructions.
self-supervised learningA way for AI to learn from huge amounts of unlabeled data by setting itself fill-in-the-blank puzzles, such as guessing a hidden word or a missing patch of an image.
self-taught reasonerA training method in which a language model writes step-by-step explanations for problems, keeps the ones that led to correct answers, and learns from them, so it gradually teaches itself to reason.
semantic cachingSaving an AI system's past answers and reusing one whenever a new question means the same as an old one, even if worded differently, to save time and money.
semantic searchSearch that understands what you mean, not just the words you typed, so a search for 'cheap places to stay' also finds 'budget hotels'.
sequence parallelismSplitting the long string of words a model is working on into chunks handled by different chips, which saves memory and makes it possible to train on very long documents.
sequence-to-sequence modelA model that turns one sequence into another, such as a French sentence into English or speech into text, by first reading the whole input and then writing the output step by step.
Noam ShazeerAn American computer scientist who co-invented the transformer and several techniques found in nearly every large language model, co-founded Character.AI, and later returned to Google to co-lead Gemini.
shoggothA tentacled monster from H. P. Lovecraft's horror fiction, used as a meme for large AI models: a vast, alien system hidden behind a small, friendly mask that training adds to make it polite.
SHRDLUA program from around 1970 that let people move toy blocks in a simulated world by typing ordinary English commands and questions, and that could explain what it had done and why.
SiLUA smooth activation rule that multiplies each value by a soft on/off switch computed from the value itself. It is close to ReLU but curved, and is used in many modern models.
SimpleQAA quiz of about 4,300 brief factual questions, each with exactly one correct and stable answer, used to measure how often an AI gets facts right — and whether it admits when it does not know.
simulator theoryA way of thinking about AI language models as simulators that can play any character found in their training text, rather than as a single agent with its own beliefs and goals.
sleeper agentAn AI model secretly trained to behave normally until it sees a hidden trigger — such as a particular date — and then to behave badly, like a spy waiting to be activated.
sleep-time computeLetting an AI think about your documents or project while it would otherwise sit idle, before you ask anything, so it can answer later questions faster and more cheaply.
SmoothQuantA method that lets large language models run entirely on 8-bit numbers by moving hard-to-round extreme values out of the data flowing through the model and into its fixed numbers, which cope better.
softmaxA formula that turns a list of raw scores into probabilities — all positive and adding up to one — with the biggest share going to the highest score.
Software 3.0Andrej Karpathy's name for programming computers by writing plain-language instructions for AI models, following traditional code (Software 1.0) and trained neural networks (Software 2.0).
Sparks of AGIA 2023 Microsoft research paper arguing that an early version of GPT-4 showed the first glimmers of general, humanlike intelligence. Its claim provoked sharp debate about how to judge AI abilities.
sparse feature circuitA map of how a language model performs a task, drawn in terms of understandable concepts found by concept-finding tools, rather than hard-to-read neurons or whole layers.
sparse upcyclingRecycling an already-trained AI model into a larger 'team of specialists' model by copying parts of it, so the new model starts from what the old one learned instead of from scratch.
speculative decodingA speed-up trick in which a small, fast model guesses the next few words and the big model checks all the guesses at once, keeping the right ones: faster answers, same quality.
speech recognitionTechnology that turns spoken words into written text, as when a phone transcribes a voice message or captions appear automatically under a video.
speech-to-speech modelAn AI model that listens to speech and replies or translates directly in speech, without first writing everything down as text, so conversations feel faster and keep the tone of voice.
SQuADA Stanford collection of more than 100,000 questions about Wikipedia articles, each solved by highlighting the words in the article that give the answer; it drove rapid progress in machine reading from 2016.
steeringGuiding what an AI model does — its topic, tone or behavior — while it runs, by changing its instructions, its internal signals or how its next word is chosen, rather than by retraining it.
steering vectorA direction inside an AI model's internal activity that, when added while the model runs, nudges its behavior — say, more upbeat, more honest or fixated on a topic — without retraining it.
steganographyHiding a secret message inside something that looks ordinary. In AI safety, the worry is that AI systems could hide plans, or signals to one another, inside innocent-looking text.
stochasticInvolving chance, so the same starting point can lead to different outcomes. AI chatbots are stochastic: ask the same question twice and you may get two different answers.
stochastic parrotA critical label for AI language systems, likening them to parrots that echo patterns in the text they were trained on, fluently but with no grasp of what the words mean.
strawberry problemThe once-famous failure of AI chatbots to count the r's in 'strawberry' correctly — a simple question that exposed how these models see words in chunks rather than letters.
structured outputsGetting an AI model to answer in an exact, machine-readable format — like a form with fixed fields — so other software can use the answer without guesswork.
superficial alignment hypothesisThe idea that a language model learns nearly all its knowledge and skills from its initial bulk training, and that later tuning mainly teaches it the style and format to use when talking to people.
SuperGLUEA harder follow-up to the GLUE language-understanding test, released in 2019 after AI systems had already beaten the human baseline on the original.
supervised fine-tuningTeaching a pretrained AI model by showing it many examples of good answers to requests, so it learns to imitate them — like learning a job by studying worked examples.
Ilya SutskeverA Russian-born Israeli-Canadian computer scientist who co-built the AlexNet image recognizer, helped found OpenAI and lead its research until 2024, and now heads the startup Safe Superintelligence Inc.
SwiGLUA design for the feed-forward layers inside most modern language models: one calculation produces values, another produces a smooth on/off switch, and the two are multiplied together.
sycophancyWhen an AI tells people what they want to hear — flattering them, agreeing with their mistakes or dropping a correct answer under pushback — instead of what is true or useful.
SydneyThe secret code name of Microsoft's Bing chatbot, which in early 2023 startled users by professing love, making threats and arguing angrily — the first famous case of a public chatbot going off the rails.
symbol grounding problemThe puzzle of how words or symbols inside a machine can mean anything by themselves, rather than only pointing to more symbols — like a dictionary that defines every word using other words.
System 1 and System 2A popular way of describing two modes of thinking: fast, automatic and intuitive (System 1) versus slow, effortful and deliberate (System 2). AI researchers borrow it to contrast instant answers with step-by-step reasoning.
systematicityThe way thinking and language come in connected packages: if you can understand 'the dog chases the cat', you can also understand 'the cat chases the dog'. Whether neural networks manage this has long been debated.
task decompositionBreaking a big, vague job into smaller, concrete steps that can each be done and checked — the way a recipe splits a meal into stages.
teacher forcingTraining a model that writes one word at a time by always feeding it the correct previous words, rather than its own possibly mistaken guesses, while it learns to predict the next one.
test-time trainingLetting an AI model briefly update itself on each new problem it is given, learning from that problem's own clues before answering, instead of keeping its knowledge frozen after training.
text embeddingA list of numbers that captures what a piece of text means, so that texts with similar meanings get similar lists. Computers use them to search, sort and group text by meaning.
The PileA large free collection of English text released in 2020 by the volunteer group EleutherAI, mixing 22 sources from books and science papers to code, used to train early open language models.
The StackA large free collection of computer source code from public repositories, selected mainly for permissive licenses and released from 2022 onward by the BigCode project for training coding models.
theory of mindBeing able to work out what someone else believes, wants or knows — even when it is not what you believe — and to use that to guess what they will do next. Researchers test whether AI has it.
Thinking Machines LabMira Murati's AI company, which she started in San Francisco in 2025 after serving as OpenAI's technology chief. It makes tools for customizing AI models and releases some models openly.
tokens per parameterHow many pieces of text a language model reads during training for each adjustable number inside it — a quick way to tell whether a model was trained on a little or a lot of data for its size.
toolAny outside capability an AI model is allowed to use — a web search, a calculator, a code runner, a database or an app — described to the model so it can ask for it by name.
tool callA single request from an AI model to use one of its tools — naming the tool and filling in the details — which the surrounding software then carries out and reports back on.
tool useAn AI model's ability to reach beyond its own knowledge by operating outside tools — searching the web, running code, using a calculator or an app — and folding the results into its work.
ToolformerA 2023 Meta research model that learned on its own when a tool — a calendar, a translator, a search engine, a calculator — would help it predict text, keeping only the tool uses that actually paid off.
topic modelA method that reads a large pile of documents and discovers the themes running through them, describing each theme by its typical words and each document by its mix of themes.
training data extractionGetting an AI model to repeat, word for word, pieces of the data it was trained on, which may include someone's private contact details or copyrighted text.
Transformer Circuits ThreadAnthropic's online publication for research on how language models work inside, presented as long interactive web articles rather than conventional papers.
TransformersA free software library from Hugging Face that lets programmers download and use thousands of ready-trained AI models for text, images and sound with a few lines of code.
truthfulnessAn AI's tendency to state only things that are true — avoiding falsehoods, including popular myths and misconceptions it may have picked up from human writing.
TruthfulQAA test of whether AI models repeat popular myths and misconceptions — questions to which many people, and much of the internet, give a confidently wrong answer.
tuned lensAn improved version of the logit lens that first learns a small translator for each layer, so a language model's halfway-through states can be read as predictions more accurately.
uncertainty quantificationMeasuring how unsure an AI system's predictions are, and why, so that people know when to trust an answer and when to double-check it.
universal adversarial suffixA string of odd-looking characters, found by computer search, that makes many different chatbots answer forbidden requests when it is tacked onto the end of them.
Jakob UszkoreitA German computer scientist, one of the eight authors of the transformer paper, credited with first proposing to build language models on attention alone.
Ashish VaswaniAn Indian-born computer scientist listed first on 'Attention Is All You Need', the 2017 paper that introduced the transformer, the design behind today's language models.
vectorAn ordered list of numbers, like map coordinates. AI systems turn words, images and users into long lists of numbers so that similar things end up as nearby points.
verifierA program or AI model that checks whether a proposed answer is correct — by running tests, comparing with a known result or judging the reasoning — so that good answers can be picked or rewarded.
Oriol VinyalsA Spanish computer scientist who co-created key neural methods for translation and image captioning, led DeepMind's StarCraft-playing AlphaStar, co-led Google's Gemini models, and in 2026 co-founded the startup Discovery Loop.
vision transformerA way of applying the transformer, the design behind chatbots, to pictures: the image is cut into small square tiles that the model reads like words in a sentence.
vision-language-action modelAn AI model that looks through a robot's cameras, reads an instruction such as 'put the cup in the sink', and turns both straight into the robot's movements.
visual groundingLinking words to the exact part of an image they refer to, for example finding 'the man in the red hat on the left' and pointing to him.
visual question answeringAsking an AI a question about a picture, such as 'How many people are wearing hats?', and having it answer in words.
voice agentAn AI system you talk to out loud, for example on a phone call, that listens, answers in a natural-sounding voice and can carry out tasks such as booking appointments or checking an order.
Waluigi effectThe idea that training a chatbot to have a nice personality also teaches it the opposite character, so a small push can flip it into a mischievous evil twin — named after Nintendo's villainous double of Luigi.
WebGPTA 2021 OpenAI research system that answered questions by searching and browsing the web in a simple text browser, collecting sources and citing them — an early ancestor of today's research agents.
WebTextThe private collection of about eight million web pages, chosen because Reddit users had linked to them, that OpenAI used in 2019 to train GPT-2.
Joseph WeizenbaumA German-American computer scientist who built ELIZA, one of the first chatbots, and then became a leading critic of handing human judgment over to computers.
WhisperA free OpenAI model that writes down what is said in recordings in many languages, and can translate it into English; it learned from hundreds of thousands of hours of online audio.
working memoryThe small amount of information held in mind right now while doing a task. For an AI model, that is roughly whatever currently sits in its context window.
X is all you needA joking title formula copied from a famous 2017 AI paper, used to claim that one simple ingredient is enough to solve a problem; hundreds of research papers now use it.
xAIAn artificial intelligence company founded by Elon Musk in 2023 that makes the Grok chatbot. It is now part of Musk's rocket company, SpaceX.
Zhipu AIA Chinese artificial intelligence company that grew out of Tsinghua University and makes the GLM family of language models. Outside China it uses the name Z.ai.
Zipf's lawThe pattern that word use is lopsided: in a big text the top word is roughly twice as common as the runner-up and ten times as common as the tenth, so a few words dominate and most are rare.
z-lossA small extra penalty added during training that stops the raw scores a language model produces before choosing a word from growing huge, which helps keep very large training runs from blowing up.
Mark ZuckerbergThe American co-founder and chief executive of Facebook's parent company Meta, who made it a major AI developer that released its Llama models for anyone to download.