AI Glossary

A step-by-step set of instructions that a computer follows to solve a problem or complete a task.

Technology that enables computers to perform tasks that typically require human intelligence, such as understanding language, recognizing images, and making decisions.

A neural network component that allows the model to focus on relevant parts of the input when generating output — the key innovation behind Transformers.

An algorithm for training neural networks that calculates how much each weight contributed to the error and adjusts them accordingly.

Bidirectional Encoder Representations from Transformers — a Google language model that reads text from both directions, greatly improving language understanding tasks.

When an AI system produces unfair or skewed outcomes, usually because the training data reflects historical inequalities or prejudices.

Extremely large datasets that are too complex for traditional data processing tools. Characterized by volume, velocity, and variety.

An AI program that simulates conversation with users, typically through text or voice. Examples: customer service bots, Siri, Alexa.

A machine learning task where the model assigns inputs to discrete categories (e.g., spam/not spam, cat/dog).

An AI assistant developed by Anthropic, known for being helpful, harmless, and honest.

A neural network architecture designed for processing visual data like images, using filters to detect patterns.

The field of AI that enables computers to interpret and understand visual information from images and videos.

The maximum amount of text (measured in tokens) that a language model can consider at one time.

Artificially expanding training data by applying transformations (flipping, rotating, cropping) to existing examples.

A collection of data used to train, validate, or test a machine learning model.

A subset of machine learning using neural networks with many layers (hence "deep") to automatically learn complex patterns from large data.

A dense numerical vector representation of data (text, images) where similar items are close together in vector space.

One complete pass through the entire training dataset during model training.

A training approach where models learn from data across multiple devices or institutions without the raw data ever leaving its source.

Further training a pre-trained model on a specific dataset to adapt it for a particular task.

A neural network architecture where two networks (generator and discriminator) compete to produce realistic synthetic data.

AI that can create new content including text, images, music, code, and video by learning patterns from training data.

Generative Pre-trained Transformer — a series of large language models by OpenAI capable of generating human-like text.

An optimization algorithm that minimizes the model's loss by iteratively adjusting weights in the direction that reduces error.

When an AI model generates confident but factually incorrect information. A significant challenge in deploying LLMs safely.

A setting that controls the training process (learning rate, batch size) rather than a value learned from data.

The process of using a trained model to make predictions on new, unseen data.

A massive AI model trained on vast text data that can understand and generate human language. Examples: GPT-4, Claude, Gemini.

The compressed internal representation that a neural network learns, where similar concepts cluster together.

A hyperparameter that controls how much the model's weights are adjusted with each training step.

Light Detection And Ranging — a sensor that measures distance using laser pulses, used in robotics and self-driving cars.

A mathematical formula that measures how wrong the model's predictions are — training aims to minimize this value.

A branch of AI where systems learn from data and improve their performance over time without being explicitly programmed.

An open standard by Anthropic that lets AI models connect to external tools and data sources in a standardized way.

AI that can process multiple types of data (text, images, audio, video) together.

The field of AI focused on enabling computers to understand, interpret, and generate human language.

A computing system loosely inspired by the human brain, consisting of interconnected nodes that process information in layers.

When a model performs very well on training data but poorly on new data because it memorized examples instead of learning patterns.

A numerical value in a neural network that is learned during training. Modern LLMs have billions or trillions of parameters.

In reinforcement learning, the strategy the agent uses to decide which action to take in each state.

Training a model on a large general dataset before fine-tuning it for specific tasks.

The input text or instruction given to an AI model to guide its output.

The skill of crafting effective prompts to get desired outputs from AI models.

A technique where an LLM retrieves relevant documents from a database before generating a response, reducing hallucinations.

A machine learning task that predicts continuous numerical values (e.g., house prices, temperature).

A type of machine learning where an agent learns by taking actions in an environment and receiving rewards or penalties.

Reinforcement Learning from Human Feedback — using human ratings of AI outputs to train more helpful and safe models.

A neural network designed for sequential data where outputs from previous steps feed back as inputs, enabling memory.

The engineering discipline that designs, builds, and programs robots — physical machines that sense, process, and act in the world.

Search that understands the meaning of a query, not just keywords — powered by embeddings and neural networks.

Simultaneous Localization and Mapping — a robotics technique for building a map of an environment while tracking position within it.

A type of ML where the model learns from labeled training data — each example has an input and a correct answer.

A unit of text (word, sub-word, or character) that language models process. "Hello world" = 2 tokens.

The process of adjusting a model's parameters using data so it can make accurate predictions.

Reusing a pre-trained model on a new but related task, saving time and data requirements.

A neural network architecture introduced in 2017 that uses attention mechanisms. The foundation of all modern LLMs.

When a model is too simple to capture the patterns in data, performing poorly on both training and new data.

ML where the model finds patterns in unlabeled data without predefined correct answers. Used for clustering and anomaly detection.

A database optimized for storing and searching high-dimensional embeddings — essential for semantic search and RAG systems.

Points you earn on AiMyGuru by completing quizzes. Higher scores earn more XP and unlock better badge ranks.

The ability of a model to handle tasks it was never explicitly trained on, using only its general knowledge.