AI Glossary

Hypothetical AI with the ability to understand or learn any intellectual task that a human being can perform.

Autonomous AI systems that perceive their environment, plan multi-step actions, and use tools to complete tasks.

The challenge of ensuring AI systems reliably pursue goals that align with human intentions and values.

Systematic errors in AI systems that produce unfair, discriminatory, or inaccurate outcomes for certain groups.

The field focused on ensuring AI systems remain beneficial, controllable, and aligned with human values as they become more capable.

The simulation of human intelligence by computer systems, enabling machines to learn, reason, and solve problems.

A neural network component that lets models dynamically focus on relevant parts of the input when generating each output token.

The algorithm for computing gradients in neural networks by propagating errors backwards through layers.

Google's 2018 bidirectional encoder model that transformed NLP by learning contextual word representations from unlabelled text.

Anthropic's training approach that uses a set of principles to guide AI self-critique and revision for safer outputs.

The maximum number of tokens an LLM can process in a single request, determining how much text it can "see" at once.

A neural network architecture using convolutional filters to extract spatial features from images or sequences.

The process of tagging raw data with labels, bounding boxes, or other metadata to create supervised learning training datasets.

A subset of machine learning using neural networks with many layers to learn complex hierarchical representations.

A generative model that learns to create data by reversing a gradual noise-addition process.

Running AI inference on local devices (phones, IoT sensors, vehicles) rather than in the cloud.

A dense numerical vector representation of data (text, images, audio) capturing semantic meaning.

Training ML models across decentralised devices without sharing raw data, preserving privacy.

Continuing training of a pre-trained model on domain-specific data to specialise it for a particular task.

A large model trained on broad data that can be adapted to many downstream tasks via fine-tuning or prompting.

A generative model architecture where two networks — a generator and discriminator — compete to produce realistic synthetic data.

AI systems that create new content — text, images, audio, video, code — by learning patterns from training data.

OpenAI's series of large decoder-only language models that established the paradigm for modern AI assistants.

Specialised processors with thousands of cores enabling the massive parallel computation required for AI training and inference.

An iterative optimisation algorithm that updates model weights by stepping in the direction that most reduces loss.

When an AI model generates plausible-sounding but factually incorrect or fabricated information.

Using a trained AI model to generate predictions or outputs on new input data.

The ability to understand, explain, and audit how an AI model arrives at its outputs.

A transformer-based AI system trained on billions of tokens of text, capable of generating, reasoning about, and transforming language.

A parameter-efficient fine-tuning method that inserts small trainable rank-decomposition matrices into model layers.

A subset of AI where systems learn patterns from data rather than following explicitly programmed rules.

The practice of streamlining the ML lifecycle — from experimentation to production deployment and ongoing monitoring.

AI systems that process and generate multiple data types — text, images, audio, and video — simultaneously.

A computing system of interconnected nodes inspired by biological brains, trained to recognise patterns.

AI models, datasets, and tools released publicly under open licences, enabling community use, inspection, and modification.

The initial phase of training a large model on massive, general datasets before task-specific fine-tuning.

The practice of crafting effective text inputs to guide LLMs toward desired outputs.

Reducing the numerical precision of model weights to decrease memory footprint and accelerate inference with minimal accuracy loss.

Grounding LLM responses by first retrieving relevant documents from a knowledge base before generating an answer.

A neural architecture that processes sequences by maintaining a hidden state across time steps.

A learning paradigm where an agent learns by taking actions in an environment and receiving reward signals.

A training technique where human preference ratings guide language model fine-tuning to produce more helpful, harmless outputs.

Learning representations from unlabelled data by creating supervisory signals from the data itself.

Sequential rounds of venture capital financing, each typically larger and at a higher valuation than the previous.

Training ML models on labelled input-output pairs to learn a function mapping inputs to target outputs.

Artificially generated data that mimics real data properties, used when real data is scarce, sensitive, or biased.

The process of splitting text into tokens (subword units) for processing by language models.

Reusing knowledge from a model trained on one task to improve learning on a different but related task.

A neural network architecture using self-attention to process sequences in parallel — the foundation of all modern LLMs.

A privately held startup valued at $1 billion or more.

Finding patterns, structure, or representations in data without labelled examples.

A database optimised for storing and querying high-dimensional embedding vectors via approximate nearest-neighbour search.

Private equity investment in early-stage, high-growth companies in exchange for equity stakes.