AI

Large language models, in plain business terms

16 April 2024 · 5 min read

The tools behind the recent wave of AI are called large language models, and the name is more helpful than it first appears. You do not need the mathematics to use them well, but a plain mental model helps you know when to trust the output and when to check it, which is most of the skill.

A very well-read pattern matcher

A useful way to picture one of these models is as something that has read an enormous amount of text and become extremely good at predicting what words should come next. That sounds modest, and yet at scale it produces writing, summaries and answers that are genuinely useful. It is closer to a fluent, well-read generalist than to a database of verified facts, and that distinction explains most of its quirks.

Why it is confident and sometimes wrong

Because the model predicts plausible language rather than looking up verified truth, it can state something incorrect with complete confidence, and even invent sources that look real. This is the single most important thing to understand. The fluency is real, but it is not the same as accuracy, which is why a human check matters wherever being wrong would cost you something.

What that means for how you use it

Give it context, because it cannot read your mind or see your files unless you show it. Lean on it for drafting, summarising and reshaping, where you can judge the result at a glance. Verify anything factual, especially names, numbers, dates and quotes. Used with that mental model, it becomes a fast, tireless assistant rather than an oracle to be trusted blindly.

Hold that picture in mind and these tools become far easier to use well: excellent for drafting and summarising, to be verified whenever facts matter.