Learn
Plain-English guides to the ideas that keep coming up in AI news. Each one explains the concept, where it goes wrong and how to judge claims about it, with links to the primary sources.
By Elliot Marsh, Editor · Updated 2 November 2026
What is an AI agent? A plain-English guide
An AI agent is a language model that can plan, use tools and act in a loop until a task is done. How agents work, where they fail and how to judge one.
5 min read
How LLM context windows work, and why bigger is not always better
A context window is how much text a language model can consider at once, measured in tokens. How it works, what it costs and why long contexts still lose detail.
5 min read
RAG explained: how retrieval-augmented generation grounds AI answers
Retrieval-augmented generation (RAG) lets a language model answer from your own documents by searching first and writing second. How RAG works and where it fails.
4 min read
What are tokens in AI? How language models read text and charge for it
Tokens are the chunks of text a language model reads and writes. How tokenisation works, why it affects AI pricing, limits and quirks, with worked examples.
4 min read
Prompting, RAG or fine-tuning? How to choose the right way to customise an LLM
Should you improve your prompts, add retrieval (RAG) or fine-tune a model? A practical guide to what each method changes, what it costs and when to use it.
4 min read
What is MCP? The Model Context Protocol explained
The Model Context Protocol (MCP) is an open standard for connecting AI assistants to tools and data. How MCP servers and clients work, and the security risks.
4 min read
Mixture of experts explained: why a trillion-parameter model can run like a much smaller one
Mixture-of-experts (MoE) models route each token to a few specialist sub-networks. What total and active parameters mean, and the trade-offs for running them.
4 min read
Open-weight vs open-source AI: what the labels really mean
Not every 'open' AI model is open source. What open weights, open-source AI and licences mean, what you can and cannot do, and how to check a model's terms.
4 min read
Why AI hallucinates, and how to reduce it
AI hallucinations are confident but false outputs from language models. Why they happen, which tasks are riskiest and practical ways to reduce and catch them.
4 min read
What is prompt injection? The security flaw at the heart of AI agents
Prompt injection tricks an AI model into following instructions hidden in emails, web pages or files. How direct and indirect attacks work and how to defend against them.
4 min read