Large language model
대규모 언어 모델
Also known as: LLM
A model trained on very large amounts of text to predict what comes next, used for summarizing, translating, writing and conversation.
The training goal is simple: given the text so far, predict the next token. Pushed to enormous data and size, that single goal produced translation and summarization without being taught them separately.
The weaknesses are just as clear. It invents plausible answers and knows nothing after its cutoff, so anything fact sensitive needs retrieval or visible sources.
In education the safer use is polishing and organizing material rather than producing answers. Sentences that reach a child are checked by a person.
- Next tokenPrediction happens over tokens, not letters.
- ReviewAnything factual gets a human check.