AI Writing
Large Language Model (LLM)
A neural network trained on very large text corpora to predict the next token in a sequence, capable of producing fluent natural-language output.
Definition
A large language model is a type of artificial intelligence built from a transformer neural network, trained on hundreds of billions of tokens of text. Its core capability is straightforward: given a prefix of text, predict what comes next. From that one capability emerges everything else — summarisation, translation, code generation, conversation, structured rewriting. The 'large' refers both to the model's parameter count (often tens to hundreds of billions) and to the scale of training data.
Why it matters
Every modern AI writing tool is a wrapper around an LLM, whether OpenAI's GPT family, Anthropic's Claude, Google's Gemini, Meta's Llama, or others. Understanding that the underlying mechanism is next-token prediction explains both the strengths (fluent, plausible prose) and the weaknesses (confident wrongness, statistically average phrasing, difficulty with truly novel reasoning). It also clarifies why prompting matters so much. A model is not retrieving facts from a database; it is sampling from a learned distribution conditioned on your input. Better-shaped prompts produce better-shaped outputs, and editorial judgment is required to verify and refine what comes back.
Examples
Strength
Drafting a structured first version of a long-form article in under a minute — useful raw material that a human then sharpens.
Weakness
Producing a list of citations that look correct but reference papers that do not exist. The model is predicting plausible text, not retrieving verified facts.
Frequently asked
Are all LLMs the same?
No. Models differ in training data, parameter count, reinforcement-learning techniques, and reasoning ability. Output quality and style vary noticeably between them.
Do LLMs understand what they write?
Not in the human sense. They model statistical patterns of language. Whether that constitutes a form of understanding is a live philosophical debate; for practical writing work, treat outputs as drafts requiring human verification.
Can LLMs learn from my edits?
Only if the underlying tool explicitly retains your edits for training, which most reputable providers do not do by default. Always check the data-usage terms of the service you use.
Related terms
Tokenization
The process of breaking text into the discrete units (tokens) a language model actually processes.
Fine-Tuning
Further training a pre-trained model on additional task-specific examples to specialise its behaviour.
Prompt Engineering
The discipline of crafting model inputs — instructions, context, examples, and constraints — to produce reliably better outputs.
Put the concept to work
Open the rewrite engine and apply this principle to a draft of your own.
Try a rewrite