AI & ML

Predictive AI analyses existing data to forecast or classify an outcome; generative AI creates new content. Here is how they differ, where each fits and why they work well together.

A diffusion model is a generative AI that creates images by learning to reverse the process of adding random noise, gradually denoising from static into a picture guided by your prompt.

A small language model (SLM) has far fewer parameters than a large one, so it is cheaper and faster to run, even on a device, trading some general breadth for efficiency.

AI governance is the policies, processes and roles an organisation uses to manage AI responsibly across its lifecycle. How it differs from responsible AI, and what it includes.

Semantic search finds results by meaning rather than exact keywords, using embeddings and vector similarity. Learn how it works and why it underpins retrieval-augmented generation.

AI needs GPUs because training and running models is mostly huge amounts of parallel matrix maths, which a GPU does thousands at a time where a CPU works more sequentially.

Machine learning is the broad field of algorithms that learn from data to make predictions; generative AI is a subset of it that creates new content. One is not an alternative to the other.

AI tokens are the small chunks of text — roughly words or word-pieces — that language models read and generate, and the unit their usage is typically measured and limited in.

A context window is the amount of text, measured in tokens, that an AI model can consider at once for a single request — covering both your input and its output.

Agentic AI describes AI systems that pursue a goal across multiple steps — planning, using tools, acting and adjusting — rather than just answering a single prompt.

The Model Context Protocol (MCP) is an open standard that connects AI models to external tools and data through one consistent interface, instead of a bespoke integration per tool.

Supervised learning trains a model on labelled examples to map inputs to known outputs, while unsupervised learning finds patterns and structure in unlabelled data on its own.

Training is the compute-intensive learning phase where a model adjusts to patterns in data, while inference is using the trained model to produce predictions or outputs on new inputs.

A neural network is a machine-learning model made of layers of interconnected nodes that learn to recognise patterns by adjusting connection weights during training.

A transformer is a neural-network architecture behind most modern AI, using an "attention" mechanism that lets a model weigh how different parts of the input relate to each other.

An AI copilot is an AI assistant embedded in a tool or workflow that suggests and drafts as you work — the human stays in control and makes the final decisions.

An AI hallucination is when a model produces fluent, confident output that is factually wrong or fabricated, because it generates plausible patterns rather than verified facts.

Multimodal AI refers to AI systems that understand and work with more than one type of data at once — text, images, audio and video — rather than a single input type.

Reinforcement learning is a type of machine learning where an agent learns by trial and error, adjusting its behaviour based on rewards and penalties to maximise reward over time.

Embeddings are numerical representations of data — text, images or other content — that capture its meaning as a list of numbers, so similar items sit close together.

A foundation model is a large AI model trained on a broad dataset that can be adapted to many tasks, rather than built for one narrow job.

Computer vision is the field of AI that enables computers to interpret images and video — identifying objects, reading text and detecting faces.

Fine-tuning is training a pretrained AI model further on your own examples so it adapts its behaviour — tone, format or task — to your specific needs.

Natural language processing (NLP) is the field of AI enabling computers to understand, interpret and generate human language, powering translation, chatbots and search.

AI agents use a language model to take actions — planning steps, using tools, and working towards a goal with some autonomy — rather than simply answering a single prompt.

A vector database stores and searches embeddings — numerical representations of meaning — so you can find items by similarity of meaning rather than exact keyword matches.

AWS, Google Cloud and Microsoft generative AI certifications compared — AI Practitioner, Generative AI Leader and AI-901: format, coding and fit.

MLOps is the set of practices for reliably building, deploying, monitoring and maintaining machine-learning models in production, applying DevOps discipline to ML’s messier lifecycle.

AI is the broad goal of machines performing intelligent tasks, machine learning is a subset that learns patterns from data, and deep learning is a further subset using multi-layered neural networks.

AI certifications are worth it as a structured, credible way to build and signal AI fluency — not as a shortcut to becoming an AI engineer. An honest look, with no invented salary claims.

A large language model (LLM) is an AI model trained on vast amounts of text to predict and generate language — letting it answer questions, summarise, translate and write in a human-like way.

Amazon Bedrock is AWS’s fully managed service for building generative-AI applications with foundation models from multiple providers through one API — no infrastructure to run.

Generative AI is a type of artificial intelligence that creates new content — text, images, code, audio — by learning patterns from existing data rather than only classifying or predicting from it.

Prompt engineering is the practice of writing and refining the instructions you give an AI model so it produces more accurate, relevant and useful output — a learnable, practical skill.

Responsible AI is the practice of building and using AI systems that are fair, transparent, safe, accountable and privacy-respecting. The core dimensions, explained plainly.

Retrieval-augmented generation (RAG) improves an AI model’s answers by first retrieving relevant information from a trusted knowledge source and supplying it as context, grounding responses in your data.