61 terms
Models & Behavior
Kinds of models, what they can take in and produce, and the settings that shape their answers.
- abstentionnounan AI system's decision to withhold a substantive answer or prediction when it lacks sufficient evidence or confidence, cannot…
- activation steeringnountrendinga method for changing an AI model's behavior during inference by modifying selected internal activations, usually without…
- AI assistantnounan AI product that helps a person with everyday tasks such as answering questions, writing, planning, or searching, usually…
- base modelnouna language model that has only been pre-trained, not yet tuned to follow instructions or chat; it continues text rather than…
- Bayesian networknouna probabilistic graphical model that represents random variables as nodes in a directed acyclic graph and describes each variable…
- chatbotnouna computer program that holds a conversation with people through text or voice
- conformal predictionnountrendinga statistical method that wraps around a predictive model to return a set or interval of possible answers with a user-chosen…
- constrained decodingnountrendinga generation technique that restricts which tokens a language model may choose at each decoding step so that the completed output…
- context windownountrendingthe maximum amount of text, measured in tokens, that a model can consider at one time, including the conversation so far and its…
- decision modelnountrendinga fast, low-cost AI model that reads text or code and returns a choice, score, or probability from a fixed set of options…
- decision networknouna graphical model that extends a Bayesian network with decision nodes and utility or value nodes so that an agent can compare…
- decision treenouna supervised machine-learning model that makes a classification or numerical prediction by sending an input through a hierarchy…
- diffusion language modelnountrendinga language model that generates text through an iterative denoising process, repeatedly turning a noisy, masked, or incomplete…
- diffusion modelnouna generative model that creates images (or audio or video) by starting from random noise and removing it step by step until a…
- discrete diffusionnountrendinga diffusion-model approach for categorical or other discrete data that learns to reverse a step-by-step corruption process…
- flow matchingnountrendinga method for training a continuous generative model by teaching it the vector field that moves samples along chosen probability…
- foundation modelnouna large model trained on broad data that can be adapted to many different tasks, serving as the base on which many products are…
- frontier modelnountrendingone of the most capable AI models available at a given time, typically the newest and most expensive to build
- full-duplex speech modelnountrendinga speech AI system that can listen and produce speech concurrently, allowing overlap, interruptions, and rapid turn-taking…
- global workspacenountrendinga limited internal system that makes selected information broadly available to otherwise specialized parts of an intelligent…
- GPTnounshort for generative pre-trained transformer: a language model that is first pre-trained on large amounts of text and then…
- hallucinationnountrendinga confident but false or made-up statement produced by an AI model, such as an invented fact, quote, citation, or source
- inverse scalingnouna pattern in which performance on a particular AI task becomes worse as model scale increases, such as when larger models in a…
- knowledge cutoffnounthe date after which a model has no information from its training data
- large language modelnountrendinga very large neural network trained on huge amounts of text to predict the next token, which as a result can answer questions…
- latencynounthe delay between sending a request to a model and getting a response, often measured as time to the first token
- long-horizon modelnountrendingan AI model designed to work autonomously on a goal through many steps and repeated attempts over hours, days, or longer, rather…
- mixture of expertsnouna model design in which many smaller sub-networks ("experts") exist, but only a few are switched on for each token, making large…
- model routingnountrendingthe automatic selection of which AI model, model tier, or model provider should handle a request, based on factors such as the…
- multimodaladjectivetrendingable to understand or produce more than one type of data, such as text, images, audio, and video
- neural operatornountrendinga neural-network model that learns a mapping from one function or field to another, often so that a physical system can be…
- open weightsadjectivetrendingdescribing a model whose trained weights are published so anyone can download, run, and modify it, even if the training data and…
- open-source modelnouna model released with its weights, code and, ideally, enough information about its training data that others can study…
- preserved thinkingnountrendingthe retention and reuse of a model's earlier reasoning blocks across later turns or tool calls, sometimes with integrity checks…
- reasoningnounin AI, a model's ability to solve problems that need several steps, such as math, planning, or debugging, often by working…
- reasoning economynountrendingthe study and optimization of how an AI system allocates tokens, computation, time, and money to reasoning so that additional…
- reasoning effortnountrendinga model or API setting that influences how much internal reasoning or test-time computation an AI system is encouraged to use…
- reasoning modelnountrendinga language model trained to work through a problem step by step before giving its final answer, trading extra time and tokens for…
- reasoning summarynountrendinga model-generated condensed account of the main steps or considerations used in reasoning, returned to a user or developer…
- samplingnounthe way a model picks each next token from its list of possible choices, which decides how predictable or varied the text is
- small language modelnouna compact language model, usually with a few billion parameters or fewer, designed to run cheaply or directly on a phone or laptop
- sparse attentionnountrendingan attention mechanism that allows each query to attend to only a selected subset of tokens or positions instead of computing…
- sparse autoencodernountrendingan autoencoder trained so that only a small fraction of its learned features activate for any one input, often used to decompose…
- speech-to-textnountechnology that converts spoken words into written text
- spiking neural networknouna neural network whose units communicate through discrete events called spikes and use the timing of those spikes as part of the…
- surrogate modelnouna model used as a computationally cheaper approximation of another model, simulation, or process; in AI-based scientific and…
- tabular foundation modelnountrendinga foundation model pretrained across many tables or synthetic data-generating tasks so that it can make predictions on a new…
- temperaturenouna setting that controls how random a model's output is: low temperature gives safer, more predictable text, high temperature…
- temporal consistencynountrendingthe degree to which an AI-generated video preserves coherent identities, object appearance, spatial relationships, lighting…
- text-to-gameadjectivetrendingof or relating to an AI system that turns a natural-language description into a playable game or interactive game prototype…
- text-to-imageadjectivedescribing AI that creates pictures from a written description
- text-to-speechnountechnology that turns written text into spoken audio, now often in highly natural or cloned voices
- text-to-vectornouna generative AI capability that turns a natural-language description into editable vector artwork made from paths, shapes, fills…
- text-to-videoadjectivedescribing AI that generates video clips from a written description
- throughputnounhow much work an AI system can handle in a given time, often measured in tokens per second
- time-series foundation modelnountrendinga foundation model pretrained on large and varied collections of time-ordered data so it can perform forecasting, classification…
- top-pnouna sampling setting that lets a model choose only from the most likely next tokens whose combined probability adds up to a chosen…
- virtual cellnountrendingan AI model that represents cellular state and predicts how a biological cell may respond to genetic, chemical, environmental, or…
- vision language modelnouna model that takes both images and text as input and can describe, answer questions about, or reason over what it sees
- voice cloningnounusing AI to copy a specific person's voice from a short recording so it can say new things
- world modelnountrendingan AI system that learns how the physical world behaves, such as how objects move and react, so it can predict or simulate what…
Advertisement