AI Glossary Accordion
A documentation template describing the details, limitations, and intended use of an AI model.
A process where the model selects the most informative data points to be labeled by a human.
A German-developed LLM known for transparency, explainability, and European language capabilities.
A step-by-step procedure for solving a problem or performing a computation.
An instruction-tuned LLaMA model developed by Stanford for educational and research use.
Identifying unusual data points that differ from the norm.
A family of AI assistants from Anthropic focused on safety and long-context understanding.
A type of AI with the ability to understand, learn, and apply knowledge across a wide range of tasks.
The simulation of human intelligence processes by machines, especially computer systems.
AI that is specialized in one specific task.
A hypothetical AI that surpasses human intelligence across all fields.
Technique allowing models to focus on relevant parts of input data, essential in transformer models.
Tools that automate the end-to-end process of applying ML to real-world problems.
A digital representation or character used to represent a user or persona, often used in AI-generated video or virtual platforms.
A transformer-based model that understands context in both directions.
A method used to train neural networks by updating weights based on error.
A Chinese LLM by Baidu, trained with large-scale data from the Chinese internet.
An AI chatbot developed by Google, based on its language models and later integrated into Gemini.
Systematic error introduced by an assumption in the machine learning process.
Techniques used to reduce bias in AI models.
A model whose internal workings are not visible or understandable.
A graphic design platform that integrates AI tools for content creation, design suggestions, and image generation.
A prompting technique where models are guided to reason through intermediate steps to reach an answer.
An AI chatbot developed by OpenAI based on the GPT series of language models.
A software application used to conduct an online chat conversation via text or text-to-speech.
Assigning inputs into predefined categories.
Grouping similar data points together in unsupervised learning.
An OpenAI model trained on code that powers GitHub Copilot.
An LLM optimized for retrieval-augmented generation (RAG) tasks in enterprise applications.
A field of AI that trains computers to interpret and understand the visual world.
When the statistical properties of target variables change over time, affecting model performance.
An AI model from OpenAI that generates images from natural language descriptions.
Techniques used to increase the amount and diversity of data.
Steps taken to clean and prepare data before training a model.
An open instruction-following model based on GPT-J, designed for commercial use.
A type of machine learning using neural networks with many layers.
A generative model that learns to create data (like images) by reversing a noise process.
Techniques to reduce the number of input variables in a dataset.
A representation of text or data in a dense vector space.
A searchable database of vector embeddings, used in semantic search and retrieval-augmented generation.
One complete pass through the entire training dataset.
AI developed and deployed in a way that respects human rights, fairness, and accountability.
AI systems designed to explain their decisions to humans.
Meta’s open-weight LLM family known for competitive performance and broad adoption.
The process of selecting and transforming variables to improve model performance.
Training machine learning models across decentralized devices or servers.
Learning from a small number of examples.
Training a pre-trained model on a specific task or dataset.
A generative language model developed by OpenAI.
Google’s next-generation AI model and platform that integrates text, image, and code understanding.
AI systems that can create new content, such as text, images, or music.
An optimization algorithm used to minimize the loss function in training.
An AI chatbot developed by xAI, a company founded by Elon Musk, integrated into the X platform (formerly Twitter).
An AI chatbot developed by Elon Musk’s xAI and integrated into the X platform.
Mechanisms to enforce ethical, safety, or behavioral constraints in AI models.
When an AI generates output that is plausible but factually incorrect or nonsensical.
An AI video generation platform that creates realistic avatars for personalized and business communication.
A collection of community-built and fine-tuned models hosted on Hugging Face.
Settings used to control the training process of a model.
The process of using a trained model to make predictions.
A structured representation of facts, entities, and relationships used for reasoning and inference.
Meta’s anticipated next-generation open LLM, expected in 2025.
A type of generative AI trained on vast amounts of text to understand and generate human-like text.
The time delay between inputting data and receiving a model’s output.
A technique for efficiently fine-tuning large language models using fewer resources.
A function that measures the error between predicted and actual outcomes.
A neural network architecture that selectively activates parts of the model for efficiency and performance.
A subset of AI that involves the use of algorithms and statistical models to enable machines to improve at tasks with experience.
A small yet powerful model from Microsoft, optimized for efficiency and reasoning.
An AI-powered image generation tool that creates artwork from text prompts.
High-performing open-weight models using sparse mixture-of-experts architecture.
A mathematical representation of a real-world process, trained to make predictions or decisions.
The process of creating a smaller model that mimics a larger, more complex one.
AI systems that understand and process multiple data types (e.g., text + image).
A field of AI that gives machines the ability to read, understand, and respond in human languages.
A network of artificial neurons that mimic the human brain’s structure to process data.
A predecessor to GPT-4, widely used for general-purpose tasks.
Advanced models known for reasoning, code generation, and multimodal input.
A model for automatic speech recognition, not an LLM but often used alongside them.
A modeling error which occurs when a model is too complex and captures noise in the data.
A large model developed by Google that powers early versions of Bard.
Crafting input prompts to effectively interact with language models.
A vulnerability where malicious prompts are embedded in input data to manipulate AI behavior.
A training method that aligns model outputs with human preferences.
Predicting continuous values from input data.
A type of machine learning where agents learn to make decisions by receiving rewards or penalties.
Combines language models with a retrieval system for better-informed responses.
A method where the model generates its own labels from raw data during training.
Refers to zero-shot, one-shot, or few-shot learning, describing how many examples a model needs to perform a task.
An open-source LLM by Stability AI, designed for creativity and code.
A type of machine learning where models are trained on labeled data.
Artificially generated data used to train AI models.
The maximum number of input and output tokens a model can process at once.
The process of splitting text into smaller pieces (tokens), often words or subwords.
The dataset used to train an AI model.
Using a pre-trained model on a new, but related problem.
A deep learning architecture that uses self-attention; forms the backbone of models like GPT and BERT.
A multilingual LLM developed by researchers at Tsinghua University.
A test to determine whether a machine can exhibit human-like intelligence.
A scenario where a model is too simple to capture the underlying pattern of the data.
Machine learning using data that has not been labeled or categorized.
The model’s sensitivity to small changes in the training dataset.
A specialized database optimized for storing and searching high-dimensional vectors.
A fine-tuned LLaMA model that performs well in dialogue and chat scenarios.
A model whose decisions and internal logic are transparent.
A family of instruction-tuned models built on top of LLaMA with improved reasoning.
Making predictions without the model having seen any examples of the task.
Elon Musk’s LLM integrated with real-time data from X (formerly Twitter).