nyelvi modell (LM)

in English: language model

basicwidespread

A computer program that has learned from vast amounts of text which words usually follow one another, and based on this, can process and write text.

In plain terms

A language model is like a very well-read friend who can finish any sentence you leave incomplete. Not because they know what you are thinking, but because they have seen so much text that they can guess with good intuition what usually follows.

This simple capability goes surprisingly far. If a program can continue a text well, it can also answer questions, translate, summarize, draft letters or write code, since these are all text-continuation tasks.

It differs from traditional software in that it does not follow pre-written rules. No one taught it grammar or spelling separately: it extracted patterns from examples itself. This makes it flexible, but also capable of confident errors.

At the core of today's chatbots and AI agents lies a very large language model; the model is the engine, while the chat interface or agent is the car built around it.

How it works

The model breaks the text into small pieces, called tokens (words, word fragments, punctuation marks), and converts each into numbers.
During training, it receives huge amounts of text and repeatedly practices the same task: guess what the next piece is. If it is wrong, its internal settings (parameters) change slightly.
After a great deal of such practice, the patterns of language leave an imprint in the parameters: much of the grammar, styles, facts and relationships.
The models behind chatbots are then further trained with sampled responses and human feedback so they answer requests helpfully rather than merely continuing the text.
During use, it calculates the probability of each continuation based on your input text, selects one, appends it, and then calculates the next segment together with it. This is how the response is built piece by piece.
There is a degree of randomness in the selection, so you do not always receive the exact same answer to the same question.

An everyday example

When your phone suggests the next word while you are typing a message, a small language model is working. When you ask a chat program to draft a polite complaint letter to your service provider, the same principle applies, only with a much larger model.

Who is affected

everyoneYou read the responses of language models daily in search engines, translators, phone text correction, and chat programs, even if you are unaware of it.

employeeThe text-based part of office work (letters, summaries, minutes, table explanations) is increasingly prepared by language models, requiring you to verify the output. → Work & HR

parent, student, teacherHomework, essays, and the way of learning change if anyone can obtain well-sounding but not necessarily correct text within minutes. → Education & Family

developerCoding agents and most new AI applications are built on language models, so their capabilities and limitations determine what can be entrusted to them. → Code

A common misconception

MythA language model is a huge database from which it looks up a ready-made, correct answer.

In realityIt does not search on its own but reformulates a probable continuation each time based on learned patterns. This is why it can write new text and also state falsehoods with complete confidence.

Limits and risks

It can generate plausible-sounding but false facts, sources, or quotes (this is called hallucination), so always verify important claims.
Its knowledge is valid only until training ends: it knows about recent events only if the program is connected to a search engine or another tool.
It can inherit biases and errors present in training texts and may perform worse in smaller languages, including Hungarian, than in English.
If the model does not run on your own machine but you use it via an online service, what you type goes to the provider. Therefore, share personal, business, or health data with it only after careful consideration.

Similar, but not the same

large language model – The large language model (LLM) is the current, massive, general-purpose type of language model; a language model can be much smaller and simpler, such as your phone's word predictor.
chatbot – A chatbot is a ready-made product with an interface, rules, and memory, while the language model is just the text-understanding and text-writing engine inside it.
AI-agent – An agent augments a language model with tools and autonomous action capabilities, whereas the model alone only generates text.

Related concepts

Broader

artificial intelligence, machine learning, Natural language processing, generative AI

Narrower

large language model, small language model

On the edge

token, transformer, hallucination, prompt, training data, multimodal model, foundation model

A brief history

The basic idea is old: Andrej Markov already examined the probabilities of letter sequences in a verse novel in 1913, and Claude Shannon showed in 1948 that English text can be imitated with the probabilities of sequences of letters and words. Most modern large language models are built on the transformer architecture published in 2017. The general public got to know them at the end of 2022 with the launch of ChatGPT.

Where things stand

By late 2026, language models are a standard part of search engines, office software, phones and customer service. The focus of development has shifted toward agents that perform tasks independently and models that handle images and audio as well as text.

Status: 2026 November

In the news

Appears in 23 Agent Age stories; the curve shows weekly occurrence in the news magazine.

Hungarian court makes new EU referral on press publications and LLM chatbotsNov 3, 2026 · Jog
Hippocratic AI unveils Polaris 6 Ultra healthcare language model at a lower priceNov 2, 2026 · Health
GitHub adds prompt-injection scanning to AI agent pull requestsOct 28, 2026 · Code
Munich court: memorisation of song lyrics in language model is relevant under copyright lawOct 23, 2026 · Jog
Radiologists and LLMs fail to spot AI-generated chest x-raysOct 15, 2026 · Health
FDA grant funds an LLM "jury" to evaluate AI-generated radiology reportsSep 28, 2026 · Health
Open source in the LLM era: maintainers face a flood of AI-generated PRs and issuesSep 21, 2026 · Code
FDA extends feedback period on generative AI medical device rules and tests language models on radiology reportsSep 21, 2026 · Health

Also known as

nyelvi modell · nyelvmodell · language model · LM · nagy nyelvi modell · LLM · large language model

Sources

Vaswani et al. (2017): Attention Is All You Need

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.