Cheat sheet: working with AI

What a language model does

A language model predicts, word by word, what is most likely to come next. It does not look anything up. That is why an answer often sounds right even when it is not.

  • A fluent answer is not evidence.
  • Ask for the source. With a source an answer can be checked, without one it cannot.

A good instruction for the AI

Role
You work in customer service for a heating manufacturer.
Context
An installer asks whether spare part ET-4711 fits series B systems. The data sheet is attached.
Task
Write a reply to the customer.
Format
Five sentences at most, polite form, and at the end the page of the data sheet you rely on.
Example
This is how we usually reply: “Hello, according to the data sheet (p. 2) ET-4711 fits series B systems.”

If the answer does not fit, say what is missing instead of starting over: “shorter”, “no jargon”, “cite the source”.

Replace confidential details

Most tasks do not need names, addresses, account or invoice numbers. Replace them with placeholders before the text goes into an AI tool. The result does not get worse.

Please cancel invoice <invoice> for <customer>, <address> and refund the amount to <IBAN>.

A folder the AI works in

workspace/            Workspace
  IDENTITY.md         Map
  CONTEXT.md          Signpost
  _config/            Rules
  sort-enquiries/     One task
    CONTEXT.md        Instructions
    input/            Input
    result/           Result
  • One task per run.
  • The agent proposes, you decide.
  • No passwords and no customer data in plain text.

Check before it goes out

  • Does the answer cite its source, and does the source really say so?
  • Do numbers, dates and names match the original?
  • Is anything missing that the recipient needs? Then ask, do not let the AI fill it in.
  • Do tone and form of address suit the recipient?
  • Is there nothing confidential in it that does not belong there?

Nexorithm, nexorithm.de/en/wissen/