Skip to content

Deterministic first: what you never let a language model do

A language model is strong at language and structure, but not at calculating, counting or looking things up. Leave that to ordinary code, and use AI to structure questions and explain results.

In short

A language model predicts words. That makes it good at understanding, summarising and structuring, but unreliable at anything with one correct answer: calculations, counting, dates, amounts. So whatever ordinary, deterministic code can calculate or look up, you never leave to a model.

What you do not leave to AI

  • Calculations. Amounts, tax, fractions, interest. A model can write out a calculation convincingly and still be wrong.
  • Date and time. What day it is, how many days until a deadline. The computer simply knows that.
  • Counting and comparing. How many invoices, which one is highest. A database can calculate that exactly from the selected data.
  • Looking up facts. You take a rate or an article number from the source, not from a model's memory.

What AI does well

  • Turning a question in plain language into a structured instruction.
  • Summarising, translating or rewriting text.
  • Extracting the right data from a messy document, which the code then checks.
  • Explaining what a result means.

How to combine them

Let the model provide the structure and the code do the arithmetic. An example from our own apps: an AI creates practice exercises, but the correct answer comes from exact calculations in the code, and every variant the AI produces is checked against them. A half-right answer does more harm than a correction.

What to watch out for

Check AI output automatically where you can, and check figures live against the source: an internal audit alone is not enough. Also keep a person in charge of decisions that matter. That is how AI becomes reliable enough for real work.

Would you like advice on using language models responsibly? Request a free advice conversation.

Updated on

AI

Personal data in the AI cloud: who is watching?

A conversation with an AI service in the cloud is not private by default. A case in Canada shows what a provider can see and share. What does that mean under the GDPR, in the United Kingdom and in Switzerland?