What a Large Language Model Actually Is
Share
You can use one of these well without knowing how it works. You cannot predict how it will fail without knowing, and the failures are the expensive part.
Direct answer
A large language model predicts what text should come next, based on patterns learned from an enormous amount of writing. That single mechanism explains why it writes fluently, why it invents facts, and why giving it source material changes everything.
Based on the AI and Automation Practitioner standard (ST1512), a Level 4 apprenticeship for people already in work. Funding rules described here are the England position and are set by the funding body.
Potentially free for you
The learner never pays. Employers potentially pay nothing either, and where the apprentice is aged 16 to 24 the government funds the whole cost even if the employer does not pay the levy.
Closing any gap. A levy transfer from another employer covers 100% of the cost. Unused transfer allowance is not carried forward at the end of the tax year, and it must be agreed before the apprentice starts.
From 1 August 2026 the government stopped adding the 10% top-up to new funds entering levy accounts, so there is less in the system than there was.
Part of our guide to AI Explained in Plain English for Work.
How it actually works
It was trained on an enormous quantity of text, learning which words tend to follow which others in which contexts.
When you ask it something, it produces the most plausible continuation, one piece at a time.
It is not looking anything up unless you have given it something to look at, or connected it to a search tool.
Everything else about its behaviour follows from those three facts, which is why they are worth twenty minutes of your attention.
Why it writes so well
Fluency is exactly what predicting plausible text produces. Grammar, register and structure are patterns and it learned them thoroughly.
It is genuinely good at form. Give it a format and it will match it more consistently than most people can.
It is also good at translation between registers, because that is a pattern transformation rather than a knowledge task.
None of that requires it to know anything, which is the point people miss.
Why it invents things
A false citation is exactly as plausible a continuation as a real one. Nothing in the mechanism prefers the true one.
It fills gaps rather than flagging them. Where a person would hesitate, it produces the shape of an answer.
This is not a bug awaiting a fix. It is a consequence of how the system works, which is why the checking habit matters more than the model version.
Names, numbers, dates, citations and legal detail are where it bites, because those are exactly the things that must be exact.
There is more on this in our guide to AI explained in plain English.
What changes when you give it a source
It stops predicting from memory and starts working from the text in front of it, which is a completely different task.
Summarising a document you pasted in is reliable. Recalling a document it may have seen in training is not.
This is why thread summaries in email work well and why questions about last year's regulation do not.
The practical rule is simple. Paste it in. Anything it has to remember, it can invent.
What the context window means
The model can only attend to a limited amount of text at once. That limit is the context window.
Exceed it and material silently drops out, usually from the middle, and the output looks complete anyway.
That is why very long documents produce confident summaries missing whole sections.
If accuracy matters on a long document, work through it in sections and check the section count yourself.
What this means under UK data protection
A language model has to receive your text to work on it. Under UK GDPR that is processing, and it needs a lawful basis like any other processing.
The practical consequence is that pasting personal data into a public tool is a processing decision, not a convenience. Your employer is the controller and carries the responsibility.
The Information Commissioner's Office has published guidance for organisations using these systems. It is free, it is written for practitioners, and it is more useful than any vendor material.
None of that makes the tools unusable at work. It makes the boundary on what goes in the thing worth writing down.
What this means for your work
Trust it on form and structure. Check it on fact, figure and name, every time.
Give it the source rather than the question wherever you can. This single habit removes most of the risk.
Treat confident tone as meaningless. It sounds equally certain whether it is right or wrong, which is the hardest thing to internalise.
Everything about safe use follows from understanding the mechanism, which is why it is the first thing worth learning.
Key takeaways
- It predicts plausible text one piece at a time.
- It is not looking anything up unless you gave it something to look at.
- A false citation is as plausible a continuation as a real one.
- Pasting in the source changes it from recall to reading.
- Confident tone carries no information about accuracy.
Frequently asked questions
What is a large language model?
A system that predicts what text should come next based on patterns learned from an enormous amount of writing. That mechanism explains both its fluency and its errors.
Why does AI make things up?
Because a false statement is just as plausible a continuation as a true one. Nothing in the mechanism prefers the truth, so gaps get filled rather than flagged.
Does it look things up?
Not unless you give it something to read or connect it to a search tool. Left to itself it produces plausible text from patterns rather than facts.
Why is it good at writing but bad at facts?
Fluency is a pattern and it learned patterns thoroughly. Facts are specific, and specifics are exactly what a plausibility engine cannot guarantee.
What is a context window?
The amount of text the model can attend to at once. Exceed it and material silently drops out, usually from the middle, while the output still looks complete.
How do I make it more accurate?
Paste in the source material rather than asking it to recall. That changes the task from remembering to reading, which it does reliably.
Will newer models stop inventing things?
They invent less, and the mechanism is unchanged. The checking habit remains the control regardless of model version.
Does confident tone mean it is right?
No. It sounds equally certain whether it is right or wrong, which is the single hardest thing about using these tools safely.
The funded route, in figures
Figures are the England position. Levy-paying employers usually draw the whole cost from the levy they already contribute, so they pay nothing extra. Employers below the £3 million pay bill threshold pay nothing for an apprentice aged 16 to 24. For an older apprentice the government funds 95 per cent, and a levy transfer from a larger employer can cover the rest. Eligibility and funding rules are set by the funding body and can change.
Where to go next
If you want this understanding built into an assessed qualification alongside applied work, the funded route covers it.
We help UK employers and employees access government-funded AI training. Training is delivered by third-party providers. We do not guarantee a place, a qualification, a job or any particular outcome. Eligibility and funding rules are set by the funding body and can change. Always check the current position before you commit. See our disclaimer.