Explain
Plain answers, before any vocabulary.
Each explainer takes one question, answers it in ordinary English with one large diagram, and then points at the research it connects to. No prior AI-safety vocabulary is assumed.
Notes: explanations, interactive work and reading. This index collects the conceptual explanations; Notes also brings together the examples, animations and articles.
01What makes an AI system agentic?The model is only one part. Instructions, context, memory, tools and an environment it can change all shape what it does, and its own actions come back as new input.02Why can history change behaviour?Earlier interaction leaves state behind. Two identical requests can produce different actions when the state behind them differs.03How do controlled experiments help measure AI behaviour?Compare conditions designed to differ in a declared factor, control measured nuisance variables where appropriate, and set the interpretation rules in advance.
New explainers are added as the research grows. For the technical version of any answer, start at ORBIT or Method.
