LLM
An LLM is an AI model trained on very large amounts of text that produces text by predicting what follows its input. Such models run behind assistants like ChatGPT, Claude and Gemini. They are strong at summarising, classifying, translating and drafting. An LLM is not a database of facts and can state wrong information fluently.
- Also called
- large language model
- In Turkish
- LLM
- Read more · Service
- Automation and AI
Why it matters
Treating an LLM as a reliable expert for every task leads to errors. It is strong at language work but can be wrong on figures, dates or current information. Where it will be used and how its output will be checked should be decided before starting.
Example
Say a firm uses an LLM to sort 100 incoming emails a day into “invoice”, “complaint” and “quote”. The model does well at routing them to the right box. But it is not asked to calculate payment amounts; that is done in the system.
Related terms
- INPINP measures the time from a visitor’s click, tap or key press until the screen visibly responds, looking at the slowest interactions of the whole visit.
- AI agentAn AI agent is a system built on a language model that picks its own steps towards a goal and uses tools to take them.
- llms.txtllms.txt is a proposed file, placed at a site’s root, that summarises the site’s most important pages in Markdown for language models.
- MCPMCP is an open protocol that lets AI applications connect to outside tools and data sources in a common way.
- PromptA prompt is the whole instruction and context given to a language model: what is asked, which information to rely on, the output’s form and what to avoid.
- APIAn API is the interface a piece of software offers to other software, defining which requests it answers and in what form.
The definitions are general information. If you think one is incomplete or wrong, write to us and we will reply within one working day. Get in touch
From our journal
From what we have written on Automation and AI; they show how the term turns up in the work.
- How to write prompts for work: 5 parts, an example, a checkThe five parts from role and context to example, a before and after on meeting notes, how to check the answer, what never goes into a public tool, and a shared prompt library for the team.
- What is RAG? AI that answers from your documentsMaking a language model search your own documents before it writes an answer: the parts of the system, where it goes wrong, testing with a fixed question set and personal data points.
- e-Fatura in North Cyprus: who must switch, how to applyWhat the Income and Tax Office’s pages and the Regulation it publishes say: paper invoices and e-Fatura stand side by side, who the obligation covers, the steps of the application, and what a business selling online has to connect.
- How to clean customer data before a CRM importTaking a copy, one row per person and a fixed set of columns, one format, duplicates, empty values, source and permission, a twenty-row test and the rules that keep the list clean.