A large language model (LLM) is a neural network trained on large amounts of text to predict the next token, which lets it generate and transform language.
How an LLM works
An LLM is trained on text to predict the next token, a word or word fragment, given what came before. Scaled up, that one objective produces a model that can summarize, translate, draft, answer questions, and follow instructions. Most current LLMs use the transformer architecture and are further tuned to follow instructions and to call tools.
What LLMs do well:
- Drafting and rewriting text in a given style.
- Summarizing long inputs, such as release notes.
- Turning loose instructions into structured output.
Where they fail:
- They can state false things fluently, often called hallucination.
- They follow instructions found in their input, which enables prompt injection.
- They have a knowledge cutoff unless given fresh data, for example through RAG.
An example
Given a changelog and the instruction "write a 280-character X post," an LLM returns a short announcement. It may also invent a feature that is not in the changelog.
Why it matters for social publishing
An LLM is the engine inside most AI agents. Its output is plausible by design, so a person should check facts before a post goes out.
SocialHandoff does not ship its own model. Brand memory rewrites use your own key for OpenAI, Anthropic, OpenRouter, or any OpenAI-compatible endpoint, and every draft passes the same review and guardrails.