Table of Contents
A practical reference guide to all things AI
This glossary provides simple, accessible definitions for artificial intelligence and agentic AI terminology. We encourage you to use this as your reference guide while learning about AI systems, from foundational concepts to advanced techniques.
As AI evolves and develops, check in with Landall Services to see what additional information and guidance we have.
Foundational AI Concepts
1. Artificial Intelligence (AI)Â
Computer systems that can perform tasks that typically require human intelligence, such as understanding language, recognising images, making decisions and solving problems.
2. Algorithm
A set of step-by-step instructions that tells a computer how to solve a problem or complete a task.
3. Model
A trained AI system that has learned patterns from data and can make predictions or generate outputs.
4. Training
The process of teaching an AI system by showing it many examples so it can learn patterns and improve its performance over time.
5. Inference
When a trained AI model is used to make predictions or generate outputs based on new input. This is the “working” phase after training.
6. Parameters
The internal settings that an AI model adjusts during training to learn patterns. More parameters generally means the model can learn more complex patterns.
7. Dataset
A collection of information (text, images, numbers, etc.) used to train or test an AI system.
Machine Learning & Deep Learning
8. Machine Learning (ML)
A type of AI where computers learn from data and improve their performance without being explicitly programmed for every situation. The system finds patterns on its own.
9. Deep Learning
A type of machine learning that uses artificial neural networks with many layers to learn complex patterns. “Deep” refers to the multiple layers of learning.
10. Neural Network
A computer system loosely inspired by the human brain, made up of connected nodes (neurons) that process and pass along information to learn patterns.
11. Supervised Learning
Training where the AI is shown examples with correct answers (labels) so it can learn to make similar predictions.
12. Unsupervised Learning
Training where the AI finds patterns in data without being told what to look for: the system discovers structure on its own.
13. Reinforcement Learning (RL)
Training where an AI learns through trial and error by receiving rewards for good actions and penalties for bad ones.
14. Transfer Learning
Using knowledge that an AI learned from one task to help it perform a different but related task.
15. Fine-Tuning
Taking a pre-trained AI model and training it further on specific data to make it better at a particular task.
16. Overfitting
When an AI model learns the training data too well and performs poorly on new data (it remembered rather than truly understood the patterns).
17. Underfitting
When an AI model hasn’t learned enough from the training data and performs poorly overall.
Large Language Models (LLMs)
18. Large Language Model (LLM)
An AI system trained on massive amounts of text that can understand and generate human-like language. Examples include GPT, Claude, and similar systems.
19. Transformers
The underlying architecture used in modern language models that processes text by paying attention to relationships between words (he “engine” that powers LLMs).
20. Token
A piece of text that the AI processes. It could be a word, part of a word, or a character. LLMs break text into tokens to understand it.
21. Context Window
The amount of text (measured in tokens) that an AI model can consider at once. Like the model’s “working memory” or how much it can see at one time.
22. Prompt
The input or instruction given to an AI model to tell it what you want it to do, i.e. the question you ask or task you request.
23. Prompt Engineering
The practice of carefully designing prompts to get better responses from AI models.
24. System Prompt
Initial instructions given to an AI that set its behaviour, personality, or role for an entire conversation.
25. Temperature
A setting that controls how creative or random an AI’s responses are. Higher temperature means more creative/varied, lower means more focused/predictable.
26. Hallucination
When an AI confidently states incorrect or made-up information as if it were true (when the model generates plausible sounding but false content).
27. Grounding
Connecting AI responses to real sources or verified information to reduce hallucinations and increase accuracy.
28. Few-Shot Learning
Teaching an AI how to perform a task by showing it just a few examples within the prompt.
29. Zero-Shot Learning
When an AI performs a task without being shown any examples, using only its general training.
30. Chain-of-Thought (CoT)
Prompting an AI to explain its reasoning step-by-step before giving a final answer, improving the accuracy on complex problems.
Agentic AI
31. Agent
An AI system that can independently take actions to achieve goals, make decisions, and interact with tools or environments. It acts autonomously rather than just responding.
32. Agentic AI
AI systems designed to act as independent agents that can plan, make decisions, use tools, and complete complex multi-step tasks with minimal human guidance.
33. Autonomous Agent
An agent that can operate independently for extended periods, making its own decisions about what actions to take to accomplish goals.
34. Tool Use / Function Calling
The ability of an AI to use external tools, APIs, or functions to accomplish tasks, such as giving the AI access to a calculator, database, or other services.
35. ReAct (Reasoning and Acting)
A framework where an agent alternates between reasoning about what to do next and taking actions, allowing it to adapt its plan as it works.
36. Planning
When an AI breaks down a complex goal into smaller steps and decides on a sequence of actions to achieve that goal.
37. Multi-Agent System
Multiple AI agents working together, often with different roles or expertise, to solve problems that would be difficult for a single agent.
38. Agent Orchestration
The coordination and management of multiple AI agents, directing them to work together effectively toward common goals.
39. Feedback Loop
When an agent observes the results of its actions and uses that information to improve or adjust its future actions.
40. Memory System
Components that allow an agent to store and recall information from past interactions, enabling it to maintain context over time.
Data & Knowledge Systems
41. Vector Database
A database that stores information as numerical codes (vectors) representing meaning, allowing AI to quickly find similar or relevant content.
42. Embedding
A numerical representation (vector) of text, images, or other data that captures its meaning. Similar items have similar embeddings.
43. Vector
A list of numbers that represent something (like text or an image) in a way that computers can mathematically compare and process.
44. Semantic Search
Searching based on meaning rather than exact keyword matches. The system understands what you’re looking for, not just the specific words you use.
45. Similarity Search
Finding items that are similar to a query item by comparing their vector representations. Used to find related content.
46. RAG (Retrieval-Augmented Generation)
A technique where an AI retrieves relevant information from external sources before generating a response, making answers more accurate and up-to-date.
47. Knowledge Base
A collection of organised information that an AI system can access and reference when answering questions or completing tasks.
48. Knowledge Graph
A network showing how different pieces of information are related to each other, like a map of facts and their connections.
49. Chunking
Breaking large documents into smaller pieces (chunks) so they can be processed, stored, and retrieved more effectively by AI systems.
50. Index
An organised catalog of data that allows for quick searching and retrieval. Like a book’s index that helps you find information fast.
AI Development & Operations
51. MLOps (Machine Learning Operations)
Practices and tools for deploying, monitoring, and maintaining AI models in production environments.
52. LLMOps (Large Language Model Operations)
Practices for managing and operating large language models, including prompt management, version control, and monitoring.
53. API (Application Programming Interface)
A way for different software systems to communicate with each other.
54. Endpoint
A specific URL or access point where you can send requests to an AI service.
55. Latency
The time delay between sending a request to an AI system and receiving a response. Lower latency means faster responses.
56. Throughput
How many requests an AI system can handle in a given time period. Higher throughput means more capacity.
57. Batch Processing
Processing multiple requests together as a group rather than one at a time (more efficient but not real-time).
58. Streaming
Delivering AI responses progressively as they’re generated, rather than waiting for the complete output.
59. Deployment
Making an AI model available for actual use in applications or services, moving from development/testing to production.
60. Evaluation Metrics
Measurements used to assess how well an AI model performs, such as accuracy, precision, recall, or user satisfaction scores.
61. Bias
Unfair prejudice in AI outputs, often reflecting biases present in training data. Can lead to discriminatory or unbalanced results.
62. Fairness
Ensuring AI systems treat all users and groups equitably, without favoring or discriminating against particular populations.
Want content like this in your inbox?
Sign up and we’ll make sure to keep you up-to-date on new technologies, trends, and promotions.
Specialised AI Techniques
63. Computer Vision
AI technology that enables computers to understand and interpret visual information from images or videos.
64. Natural Language Processing (NLP)
The field of AI focused on helping computers understand, interpret, and generate human language in text or speech form.
65. Natural Language Understanding (NLU)
A subset of NLP focused specifically on comprehending the meaning and intent behind language, not just identifying words.
66. Natural Language Generation (NLG)
Creating human-like text or speech from data or structured input. The AI writes or speaks in natural language.
67. Sentiment Analysis
Determining the emotional tone or attitude expressed in text (positive, negative, neutral).
68. Named Entity Recognition (NER)
Identifying and categorising important elements in text, such as names of people, places, organisations, dates, and other specific items.
69. Classification
Sorting items into predefined categories. For example, identifying whether an email is spam or marking customer feedback as positive/negative.
70. Regression
Predicting a numerical value based on input data. For example, estimating house prices or forecasting sales numbers.
71. Clustering
Using data to naturally group similar items together without predefined categories.
72. Anomaly Detection
Identifying unusual patterns or outliers that don’t fit normal behavior, used for fraud detection, quality control, and spotting problems.
73. Recommendation System
AI that suggests items users might like based on past behaviour or preferences.
74. Generative AI
AI systems that create new content (text, images, music, code, etc.) rather than just analysing existing content.
75. Multimodal AI
AI that can understand and work with multiple types of data (text, images, audio, video) together, not just one type.
Additional Important Terms
76. Foundation Model
A large AI model trained on broad data that can be adapted for many different tasks.
77. Pre-training
The initial phase of training where a model learns general patterns from large amounts of data before being specialised for specific tasks.
78. Attention Mechanism
A technique that helps AI models focus on the most relevant parts of input when processing information.
79. Self-Attention
When different parts of the input pay attention to each other to understand relationships and context (key component of transformer models).
80. Tokenisation
Breaking text into smaller pieces (tokens) that an AI model can process.
81. Quantisation
Reducing the precision of numbers in a model to make it smaller and faster while maintaining reasonable performance.
82. Pruning
Removing unnecessary parts of an AI model to make it more efficient without significantly hurting performance.
83. Distillation
Training a smaller, simpler model to mimic a larger model’s behavior, creating a compact version that’s faster and cheaper to run.
84. Edge AI
Running AI models directly on local devices (phones, cameras, sensors) rather than in the cloud, enabling faster responses and better privacy.
85. Synthetic Data
Artificially generated data created by AI or algorithms, used for training when real data is limited, expensive, or sensitive.
86. Data Augmentation
Creating modified versions of existing training data (like rotating images or rewording text) to increase the variety of examples.
87. Benchmark
A standard test or dataset used to compare the performance of different AI models.
88. Hyperparameter
Settings chosen before training that control how an AI learns, such as learning rate or number of layers.
89. Backpropagation
The process of adjusting a model’s internal settings by working backward from errors to improve future predictions.
90. Loss Function
A measure of how wrong a model’s predictions are. Training aims to reduce this loss, making predictions more accurate.
91. Epoch
One complete pass through the entire training dataset. Models typically train for many epochs to improve performance.
92. Batch Size
The number of training examples processed together before updating the model, which affects training speed and memory usage.
93. Learning Rate
Adjustments a model makes during training. Too high and it misses optimal settings; too low and training takes forever.
94. Gradient Descent
The optimisation method used to reduce errors during training by gradually adjusting parameters in the direction that reduces loss.
95. Activation Function
A mathematical function in neural networks that decides whether a neuron should activate. Adds non-linearity to enable complex learning.
96. Convolutional Neural Network (CNN)
A type of neural network especially good at processing grid-like data such as images, commonly used in computer vision.
97. Recurrent Neural Network (RNN)
A neural network designed to work with sequential data by maintaining memory of previous inputs, useful for time series and language.
98. LSTM (Long Short-Term Memory)
A specialised RNN that can remember information over long sequences better than standard RNNs, good for language and time-based tasks.
99. GAN (Generative Adversarial Network)
Two neural networks competing with each other – one creates fake data, the other tries to detect fakes. This competition creates realistic outputs.
100. Diffusion Model
A generative model that creates new content by gradually removing noise from random input, used in modern image generation systems.
101. Federated Learning
Training AI models across multiple devices or locations while keeping data local. Improves privacy by not centralising sensitive information.
102. Active Learning
A training approach where the AI identifies which examples would be most helpful to learn from and requests labels for those specific cases.
103. Ensemble Learning
Combining predictions from multiple models to get better results than any single model.
104. A/B Testing
Comparing two versions of an AI system by showing different versions to different users and measuring which performs better.
105. Cold Start Problem
The challenge when a system has no historical data about new users or items, making it difficult to provide personalised recommendations.
106. Explainability / Interpretability
Understanding why an AI made a particular decision or prediction; making AI’s reasoning transparent and understandable to humans.
107. Black Box
An AI system whose internal workings are difficult or impossible to understand, even though we can see its inputs and outputs.
108. Ethical AI
Development and deployment of AI systems in ways that are fair, transparent, accountable, and respect human rights and values.
109. Responsible AI
Ensuring AI systems are developed and used in ways that benefit society, reduce harm, and align with ethical principles and regulations.
Next Steps
This glossary provides foundational understanding of AI and agentic AI concepts. As these technologies evolve rapidly, new terms and techniques will continue to emerge.
For advice on AI security, governance or compliance, as well as new developments that could impact you or your business, talk to Landall Services today.
