Claude is a large language model (LLM) built by Anthropic. That phrase carries a lot of weight, so let's unpack it. 'Large language model' means Claude was trained on enormous amounts of text and learned to predict what comes next — which, at scale, produces a system that can reason, write, code, analyse, and converse at a high level. 'Built by Anthropic' means it was trained with a specific safety methodology called Constitutional AI, which shapes not just what Claude can do but how it behaves when it isn't sure, when it's asked to do something harmful, and when it disagrees with you.
Claude is not a search engine. It doesn't retrieve pages from the internet when you ask a question — it generates an answer from patterns in its training data. Claude is not a database. It can't look up your records, query your files, or remember what you told it last week. Claude is not a calculator. It reasons through problems in language, which means it can make arithmetic mistakes that a spreadsheet never would. Understanding what category of tool Claude belongs to is the most important thing you can learn in this program.
Most AI systems are trained using Reinforcement Learning from Human Feedback (RLHF): human raters score outputs, and the model learns to produce outputs the raters prefer. Anthropic extended this with Constitutional AI (CAI) — a process where the model also critiques and revises its own outputs against a written set of principles (the 'constitution'). This means Claude's values aren't surface guardrails bolted on after training — they're built into the model's generative process itself.
Anthropic trains Claude around three core values. These aren't marketing language — they describe real, testable properties of how the model behaves.
Knowing what Claude cannot do is as valuable as knowing what it can. The most common disappointment with Claude traces to expecting search-engine behaviour (current facts), database behaviour (your own data), or calculator behaviour (guaranteed arithmetic). Once you have the right model, you can design around the gaps.