Your CEO said "AI" at the town hall. LinkedIn won't shut up about it. Now a customer's procurement form asks whether you "use AI in your quality processes", and you nod along without knowing what's inside the box.
Five minutes, no computer science, tied to your real work.
The bookworm
LLM stands for Large Language Model. It is the engine inside ChatGPT, Claude, Gemini, Copilot, all of them. It has read a huge amount of text and learned how words and ideas connect. When you type, it predicts what should come next, word by word, from everything it has read.
The picture I use: an employee who spent ten years reading every machining textbook, every ISO standard, every quality manual, and never once stood on a factory floor. Enormous reading, zero hands-on, so useless alone and brilliant next to you. You bring the floor experience; it brings the reading.
What that means at your desk
It can process your WIP export, draft your SOP, and build your slides. It cannot verify your numbers, know which customer hates airfreight, or make a business decision. Hold that boundary and the tool behaves.
And because it predicts language, phrasing decides everything. "Analyse this data" gets you a horoscope. "Show me which customers had declining lot volumes year-over-year" gets you something you can carry into a meeting.
| LLM | Made By | You'll See It In | Good For |
|---|---|---|---|
| GPT-4 / GPT-4o | OpenAI | ChatGPT, Copilot | General tasks, writing, code |
| Claude | Anthropic | Claude.ai, Claude Code | Long documents, data analysis, file processing |
| Gemini | Google Workspace, Bard | Google integration, search | |
| Llama | Meta | Open-source tools | Self-hosted, privacy-sensitive use |
What an LLM is NOT
- Not a database: it doesn't store your files or remember past conversations.
- Not a search engine: it generates from patterns, not live lookups.
- Not infallible: it can sound confident and still be wrong. Verify anything that matters.
Ah Hock's summary
It is a pattern engine that read everything and experienced nothing. Don't ask it to be the engineer. Treat it as the fastest bookworm you ever hired, then check its work the way you'd check a new operator's first part.
