AlignmentMaking AI systems pursue the goals and values their developers and users intend.Intermediate
Computer useAgents operating software through screenshots, mouse and keyboard, like a person.Intermediate
Constitutional AITraining with a written set of principles and AI feedback to shape model behaviour.Advanced
Context engineeringCurating what an agent sees at each step: summarising, pruning and loading information just in time.Advanced
Elo ratingA rating derived from head-to-head comparisons, used by preference leaderboards.Intermediate
Large language model (LLM)A neural network trained on vast text to understand and generate language.Novice
MCP (Model Context Protocol)An open standard for connecting AI agents to tools, data and prompts.Intermediate
Mixture of experts (MoE)An architecture that routes each token to a few of many expert sub-networks.Advanced
MultimodalAble to handle several kinds of input or output, such as text, images, audio and video.Beginner
Photo finishOn AIModelRace: a race where the leader's margin is under 4% of the category's meaningful span.Novice
Post-trainingTraining after pre-training that makes a model helpful, safe and good at reasoning.Intermediate
Prompt injectionMalicious instructions hidden in content an AI reads, trying to hijack it.Intermediate
Reasoning modelA model trained to think at length before answering, improving accuracy on hard problems.Beginner
Retrieval-augmented generation (RAG)Fetching relevant documents and adding them to the prompt so answers are grounded.Intermediate
RLHFReinforcement learning from human feedback: training models toward responses people prefer.Intermediate
System promptStanding instructions that set a model's role, rules and context for a conversation.Beginner
Test-time computeExtra computation spent while answering, such as longer reasoning, to improve results.Advanced
Vector databaseA database optimised for storing embeddings and finding similar ones fast.Intermediate