跳过主要内容
全部术语
How Engines Retrieve & Rank Sources更新于

Large language model (LLM)

又称: LLM, foundation model, language model

一句话解释

A large language model is a neural network trained on very large amounts of text to predict likely continuations of language. That single capability produces answering, summarising and writing. It is the component underneath ChatGPT, Claude, Gemini and the AI features inside search engines.

What the model actually does

An LLM predicts the next piece of text given everything before it. Trained at sufficient scale, that objective produces something that reads as reasoning, summarising and answering — but the mechanism underneath is prediction over patterns in language, not lookup in a database.

Two consequences follow, and both matter commercially:

  • It has no notion of truth, only of likelihood. This is the root of hallucination.
  • It cannot cite what it does not have. A model answering from memory has no source to attach, which is why grounded retrieval was added.

Model versus product

Almost everything a brand can influence sits in the product layer, not the model. The model is frozen after training; the product searches the web, retrieves pages, and passes them to the model at answer time.

So when someone asks how to "get into the training data", the more useful question is how to be retrieved — see grounding and LLM SEO.

The knowledge cutoff, and why it matters less than it sounds

Every model has a date beyond which it knows nothing. This sounds like it should dominate what an assistant says about your business, and increasingly it does not, because the products search the live web for anything current.

Where the cutoff still bites is in defaults: an assistant asked a general question without retrieving may answer from stale memory, describing your product as it was two years ago. That is a real problem and it is fixed by being clearly and currently described on the sources that do get retrieved.

常见问题

What is the difference between an LLM and a chatbot?
The model is the engine; the chatbot is the product built around it. ChatGPT is a product that adds a conversation interface, web search, memory and safety systems on top of an underlying model. The distinction matters because most of what affects your visibility lives in the product layer, not the model.
Do LLMs know about my business?
Only if you appeared in the training data or the product retrieves your pages when asked. For most businesses the second is what matters — the model itself likely knows little, but the product searches the live web before answering.
What is a knowledge cutoff?
The date after which a model's training data ends. Anything that happened later is unknown to the model itself, which is why products bolt on live retrieval. It also means an ungrounded answer about a fast-moving topic may be confidently out of date.

相关术语

  • Grounding (AI)

    Grounding is the practice of connecting a language model's answer to specific external sources retrieved at answer time, rather than relying on what the model memorised during training. A grounded answer can cite where each claim came from, which is what makes citations — and therefore AI visibility — possible at all.

  • AI hallucination

    An AI hallucination is a statement produced by a language model that is fluent, confident and false. It happens because models predict likely text rather than retrieve verified facts, so a plausible-sounding answer is generated with the same confidence as a correct one.

  • Retrieval-augmented generation (RAG)

    Retrieval-augmented generation is a technique where a system retrieves relevant documents and places them in a language model's context, so the model answers from those sources rather than from memory alone. It is how AI answers come to have citations, and how live information reaches a frozen model.

  • LLM SEO

    LLM SEO is the practice of influencing what large language models say about your brand — getting named in their answers and cited in their sources. It is a synonym for generative engine optimization, emphasising the model rather than the product built on top of it.

相关指南