Foundations · MLD-0625

Large language model

A language model trained at substantial parameter and data scale, often usable across many text tasks.

Expanded guide · editorial review 2026-11-01. Original definitions, examples, notes and citations are preserved below. New examples are illustrative, not benchmark results.

Prerequisites

How it works

A language model assigns probabilities to token sequences. Many generative LLMs learn next-token prediction during pretraining, then receive additional task or preference training. A prompt supplies context during inference; it does not normally update the model weights. Large is a relative description, with no universal parameter cutoff.

Worked example

Give a model two examples of transforming a meeting note into an action list, then supply a third note. It may continue the demonstrated pattern without a training update. This illustrative use of examples in context differs from fine-tuning a model on a dataset.

A useful equation

An autoregressive model factorizes P(tokens) = product_t P(token_t | tokens_before_t). Decoding selects or samples from those conditional distributions.

What to test

Use unseen prompts with ambiguous instructions, missing facts and conflicting evidence. Measure task success and unsupported claims separately. Compare prompts under the same model version and generation settings.