AI explained

The words you keep hearing about AI, explained in plain English with an example you can try.

What is AGI (artificial general intelligence)? What the labs mean, and where they disagreeWhat AGI means, how OpenAI, Google DeepMind, Anthropic and ARC Prize define it differently, what their leaders predicted in 2026, and how to judge AGI claims at work.
What is AI hallucination? Why AI makes things up and how to catch itWhat AI hallucination is, why ChatGPT, Claude and Gemini still make things up in 2026, the jobs where it bites hardest and five checks that catch it before it costs you.
What is AI safety? What it covers, why it matters at work and what you can doWhat AI safety means, in plain English: misuse, malfunction and systemic risk, what the International AI Safety Report 2026 found, a 15-minute check to try and the myths.
What is MCP (Model Context Protocol)? The plug that connects AI to your toolsWhat the Model Context Protocol (MCP) is, how hosts, clients and servers work, where Claude, ChatGPT, Zapier and Make use it, and how to connect one safely at work.
What is RAG? Retrieval-augmented generation explained for non-engineersWhat retrieval-augmented generation (RAG) is, how it works in four steps, where you already use it in Gemini Notebook, Perplexity and Vibe, and when it beats fine-tuning.
What is a GPU for AI? Why AI runs on graphics chips, and what it means for youWhat a GPU is and why AI runs on it, in plain English: parallel maths, training versus inference, NVIDIA Vera Rubin, what it means for your AI tools and a check to try.
What is a context window? AI memory limits explained, with 2026 sizesWhat an AI context window is, how big it is in ChatGPT, Claude, Gemini and other models in November 2026, why bigger is not always better, and how to work within it.
What is a custom GPT? What it was, why it retires on 11 December 2026 and what replaces itWhat a custom GPT is, why OpenAI retires custom GPTs on 11 December 2026, how to migrate one to a ChatGPT plugin, what does not transfer and how Gemini Gems compare.
What is a diffusion model? How AI turns noise into images, explained simplyWhat a diffusion model is, in plain English: how it turns random noise into an image step by step, why it matters for work images, an exercise to try and the myths.
What is a prompt in AI? Meaning, parts of a good one and examplesWhat a prompt is in AI, the five parts of a prompt that works, how prompts differ from skills and system prompts, and a before-and-after example you can try now.
What is a reasoning model? How AI "thinking" works and when to use itWhat a reasoning model is, how hidden thinking tokens and effort levels work in ChatGPT, Claude, Gemini and DeepSeek, when the extra time pays off and when it does not.
What is a system prompt? The hidden instructions behind every AI assistantWhat a system prompt is, how ChatGPT, Claude and Gemini use one, how it differs from your prompt and custom instructions, and how to write a good one for your team.
What is a token in AI? How tokens work, with real numbersWhat a token is in AI, how many words a token is, how images and audio are counted, why tokens decide your limits and costs, and how to check a token count today.
What is an AI API? How apps talk to ChatGPT, Claude and Gemini behind the scenesWhat an AI API is, how it differs from a ChatGPT or Claude subscription, how token billing and API keys work, and how to make a first request safely at work.
What is an AI agent? A plain-English explanation with real examplesWhat an AI agent is, how it differs from a chatbot and a workflow, which agents ChatGPT, Claude, Copilot and Gemini offer in November 2026, and a safe first task to try.
What is an AI benchmark? How to read model scores without being fooledWhat an AI benchmark is, in plain English: how tests like Humanity's Last Exam and Terminal-Bench work, why scores mislead, and how to build a 10-task test of your own.
What is an LLM? Large language models explained in plain EnglishWhat a large language model (LLM) is, how it learns, which LLMs power ChatGPT, Claude and Gemini in November 2026, and a ten-minute test you can try today.
What is an embedding in AI? A plain-English explanation with an example you can tryWhat an embedding is in AI, explained in plain English: how meaning becomes numbers, why it powers search and RAG at work, an exercise to try and the myths to drop.
What is an open-source AI model? Open weights vs open source, explained in plain EnglishWhat an open-source AI model is, how it differs from open weights under the OSI definition, and why it matters at work, with DeepSeek, GLM and Mistral examples.
What is computer use in AI? When AI clicks and types for you, explainedWhat computer use means in AI: how Claude and other agents see your screen, click and type, where it helps at work, a safe first task to try and the risks to know.
What is deep research in AI? How ChatGPT, Gemini, Claude and Perplexity write cited reportsWhat AI deep research is, how it plans, searches and writes a cited report, which tools offer it on free and paid plans in 2026, and how to check its work before use.
What is fine-tuning in AI? What it is, when it helps and what to do insteadWhat fine-tuning an AI model means, how it differs from prompting and RAG, why OpenAI is winding down its fine-tuning platform, and what most teams should use instead.
What is generative AI? A plain-English definition with examples for workWhat generative AI is, how official bodies define it, what it can create in 2026 (text, images, video, music, voice), the main risks NIST lists and a first task to try.
What is inference in AI? Why every answer costs time and money, explained simplyWhat inference means in AI, in plain English: how it differs from training, why speed and limits depend on it, how effort settings change cost, and a test you can try.
What is multimodal AI? How AI reads images, audio and video, with examplesWhat multimodal AI is, which models handle images, audio and video in November 2026, which ones are text only, how media is counted in tokens and three tests to try.
What is prompt injection? The number one AI security risk, explained simplyWhat prompt injection is, in plain English: direct and indirect attacks, why OWASP ranks it LLM01, how it hits AI agents at work, a safe test to try and real defences.
What is temperature in AI? The randomness setting, and why the newest models drop itWhat temperature means in AI models, how low and high values change answers, why Claude Opus 5.5 removed it and Gemini 4 ignores it, and what to use instead at work.
What is text-to-speech (TTS)? How AI voices work and how to use them at workWhat text-to-speech is, in plain English: how AI voices like Eleven v4 and Gemini 3.8 Flash TTS work, where they help at work, a test to try and the rules on cloning.
What is the EU AI Act? The rules, the new timeline after the AI Omnibus, and what it means at workThe EU AI Act in plain English: the four risk levels, banned practices, the timeline the AI Omnibus changed in 2026, AI literacy duties and what your team should do now.
What is vibe coding? Building software by describing it, and where it goes wrongWhat vibe coding means, who coined it, the tools people use for it (Lovable, Bolt, Replit, v0, Claude Code), what it is good for at work and the risks to check first.