Tags
AI, Artificial Intelligence, chatgpt, llm, technology, terms
Martyn Rhisiart Jones
A Coruña, 6th December 2025
Here is the full content of Popular AI Terms Explained, transcribed as clean, readable text tables, organised by section.
| Section | Term | Explanation |
|---|---|---|
| Core Definitions | Artificial Intelligence (AI) | Systems that carry out tasks typically requiring human cognition: reasoning, learning, perception, decision-making |
| Machine Learning (ML) | AI subset where systems learn patterns from data rather than using explicit programmed rules | |
| Deep Learning (DL) | ML using multi-layered neural networks to learn hierarchical representations from large datasets | |
| Neural Network | Using a small amount of labelled data and a large amount of unlabeled data | |
| Learning Paradigms | Supervised Learning | Computational model using interconnected nodes (neurons) organised in layers to process information |
| Unsupervised Learning | Training on unlabeled data to discover hidden patterns or groupings | |
| Reinforcement Learning (RL) | Learning by trial-and-error from rewards and penalties based on actions | |
| Semi-Supervised Learning | A transformer-based model trained on vast text to understand and generate language | |
| Self-Supervised Learning | Generating labels from the data itself (common in modern large language models) | |
| Transfer Learning | Reusing a pre-trained model on a new but related task | |
| Few-Shot Learning | Learning from very few examples (typically 1–10) | |
| Zero-Shot Learning | Performing tasks without task-specific training examples | |
| Model Architecture & Training | Transformer | Architecture using self-attention mechanisms; foundation of modern LLMs |
| Large Language Model (LLM) | AI subset where systems learn patterns from data rather than using explicitly programmed rules | |
| Pre-training | Initial training on massive unlabeled data to learn broad patterns | |
| Fine-tuning | Further training a pre-trained model on smaller, task-specific data | |
| Parameters | Learnable weights in a model (billions in large models) | |
| Hyperparameters | Configuration settings fixed before training (learning rate, batch size, etc.) | |
| Training Data | Dataset used to teach a model patterns and knowledge | |
| Epoch | One entire pass through the entire training dataset | |
| Batch Size | Number of samples processed before updating model parameters | |
| Gradient Descent | A system that perceives the environment, reasons, and takes actions toward goals | |
| Generative AI | Generative AI | AI that creates new content (text, images, audio, video, code) |
| Prompt | Input text/instruction that directs a generative model’s output | |
| Token | Basic unit of text processing (word, subword, or character) | |
| Context Window | Maximum token length a model can consider at once | |
| Temperature | Parameter controlling output randomness (lower = more deterministic) | |
| Diffusion Model | A prompting technique that elicits step-by-step reasoning | |
| AI Agents & Systems | AI Agent | A generative model that progressively adds then removes noise to create data |
| Agentic AI | Autonomous agents that plan, use tools, and complete complex tasks | |
| Tool Use | AI’s ability to invoke external functions (search, calculators, code execution) | |
| Chain-of-Thought (CoT) | Multiple specialised agents collaborating to solve problems | |
| ReAct | Reasoning + Acting loop (think → act → observe → think) | |
| Multi-Agent Systems | Combining the retrieval of external knowledge with generation | |
| NLP & Language | Natural Language Processing (NLP) | AI focused on understanding, interpreting, and generating human language |
| Embedding | Dense vector representation of words or sentences in semantic space | |
| Tokenization | Splitting text into tokens for model processing | |
| RAG (Retrieval-Augmented Generation) | Model memorises training data instead of generalizing | |
| Vision & Multimodal | Computer Vision | AI that interprets and analyzes visual information from images/videos |
| Multimodal AI | Models that process and combine multiple data types (text, image, audio) | |
| OCR | Optical Character Recognition | |
| CLIP | Model that connects images and text in a shared embedding space | |
| Model Behavior & Problems | Hallucination | Model generating plausible but factually incorrect or fabricated content |
| Overfitting | Model memorises training data instead of generalising | |
| Underfitting | The model is too simple to capture patterns | |
| Bias | Standardised test for comparing model performance | |
| Prompt Injection | Malicious inputs designed to override a model’s instructions | |
| Evaluation & Metrics | Benchmark | An optimisation algorithm that iteratively adjusts parameters to minimize loss |
| Accuracy | Proportion of correct predictions | |
| Precision / Recall / F1 | Metrics for classification, especially imbalanced datasets | |
| Perplexity | Measures how well a language model predicts text (lower = better) | |
| Ground Truth | Verified correct answer used for training or evaluation | |
| Safety & Ethics | Explainable AI (XAI) | Architecture where different sub-networks specialise in different inputs |
| Red Teaming | Adversarial testing to identify vulnerabilities | |
| Constitutional AI | Training approach using principles to guide behavior | |
| Guardrails | Constraints preventing harmful or undesired outputs | |
| Model Card | Documentation of a model’s capabilities, limitations, and intended use | |
| Infrastructure & Deployment | API | Application Programming Interface for interacting with models |
| Inference | Running a trained model to generate outputs | |
| GPU / TPU | Graphics/Processing Units; hardware accelerating training and inference | |
| Quantization | Reducing model precision (e.g., 32-bit → 8-bit) to decrease size and speed inference | |
| Edge AI | Running AI models locally on devices rather than cloud servers | |
| Advanced Concepts | Latent Space | Compressed representation space where data is manipulated |
| Emergent Abilities | Capabilities appearing in large models that were not explicitly trained for | |
| Scaling Laws | Predictable relationships between model size, data, and performance | |
| In-Context Learning | Model adapts to tasks from examples provided in the prompt | |
| Mixture of Experts (MoE) | Architecture where different sub-networks specialize in different inputs | |
| AGI | Artificial General Intelligence (hypothetical AI matching human-level reasoning across all cognitive tasks) | |
| ASI | Artificial Superintelligence (theoretical AI surpassing human intelligence in all areas) |
Discover more from GOOD STRATEGY
Subscribe to get the latest posts sent to your email.
GOOD STRATEGY 