Bolds the start of each word so you can scan the text faster.
The words you meet across the manual, explained in two sentences. For each term you also get a button that generates a prompt to paste into your AI to dig deeper.
In alphabetical order.
A label that lumps together programs able to do things that used to require a person: understand a piece of text, recognize a voice, suggest the next word. There are two big families: single-task AI (spam filter, voice recognition, autocomplete) and conversational AI (ChatGPT, Claude, Gemini). The manual focuses on the second.
Everything the AI has “in front of it” while it generates the answer: your question, the previous answers in the same chat, any documents you have given it. The more useful context you provide, the better it answers. There is a limit to how much context it can re-read at once (called the context window): in recent products it is huge, but in very long conversations it can run out.
The unique “signature” of an academic article, a string that starts
with 10. (e.g., 10.1016/j.jhealeco.2019.102225). Pasted into
doi.org/<DOI> it opens the original paper: if it doesn’t resolve,
the citation is almost always made up. Almost every scientific
article published after 2000 has one.
The business model of ChatGPT, Claude, Gemini and similar products: a free basic version and paid versions that are more capable. The companies make money from subscribers and from investors who are betting on the future of the product.
When the AI produces text that sounds true but isn’t grounded in any real fact (made-up dates, citations that never existed, books that don’t exist). It isn’t a glitch: it is an effect of the “pick the most probable word” mechanism, not the truest word.
The technical term for conversational AIs. Large because it has read a huge amount of text during training, and because the mathematical program that runs it is enormous. When you hear “LLM” referring to ChatGPT/Claude/Gemini, this is what they mean.
By default, the AI does not remember you between one chat and the next: every new conversation starts from zero. Within the same chat it remembers everything you have said, because it re-reads it every time as part of the “question” it bases the next answer on. Some products add an optional “memory” feature that persists across chats too: you can turn it on and off.
The mathematical program (a huge collection of formulas) that powers an AI. When people say ChatGPT, Claude, Gemini “use different models”, they mean each product is built on a mathematical program that’s been put together differently, with different traits: speed, answer quality, cost to run.
An academic journal where every article, before being published, is read and evaluated anonymously by other experts in the field (the peers). It is the standard filter to tell an academic paper apart from a high-level blog: serious, but not reviewed. In a thesis or research paper, “peer-reviewed source” is what supervisors ask for by default.
What you write to the AI to get an answer: a question, a request, an instruction, a whole paragraph of context. The word would translate to “request” or “message”, but “prompt” is the term everyone uses now. The clearer the prompt, the better the AI answers.
A study method where you review a card right when you’re about to forget it: first after one day, then three, then a week, then a month. Each pass costs less than the previous one, and the memory actually sticks. It’s the principle behind flashcard tools like Anki and Quizlet, and one of the most reliable levers for exam preparation.
The phase in which an AI “learns” by reading huge amounts of text: books, web pages, conversations, code. From all that material it picks up patterns about how sentences are built and how arguments are made on a topic.