LLM mechanics (tokens, embeddings, context)
In plain words
A large language model (LLM) reads and writes text in small pieces called tokens: whole words, parts of words or punctuation. Inside the model, each token is turned into a list of numbers called an embedding, which places words with similar meanings close together. The context window is everything the model can "see" at once: your instructions, any documents you paste in, and the conversation so far. Anything outside that window does not exist for the model.
Healthcare example (illustrative)
Imagine an intensivist pastes a long ICU stay, many days of notes, into a chatbot and asks for a summary. If the notes are longer than the context window, the earliest days may be cut off silently, and the summary will miss the admission diagnosis. A shorter, well-chosen extract often gives a better summary than everything at once.
A little more technical
The model works by predicting the most likely next token, one at a time, given the tokens already in its context. Embeddings are what let it treat "myocardial infarction" and "heart attack" as close in meaning even though the words differ. Because the model predicts rather than looks up, it can produce fluent text that is wrong, so every output still needs human verification before it informs clinical judgement.
Don't learn this yet
Skip the maths of attention and transformer layers, and skip training your own model. Knowing that text becomes tokens, tokens become numbers, and the model only sees its context window is enough for now.