From rules to learning
Key takeaway: Classic software follows rules a person wrote. Modern AI learns patterns from examples instead.
Two ways to build software
Traditional software is a list of instructions: if the invoice total is over 1,000, ask for approval. A person decides every rule in advance, and the program can only handle situations that person imagined.
Machine learning flips this. You show the system thousands of examples of inputs and correct outputs, and it adjusts millions of internal numbers until its guesses match the examples. Nobody writes the rule; the rule emerges from the data.
Why that matters day to day
- AI is great at fuzzy tasks (summarise this, rewrite that, classify these) where rules are hard to write.
- AI is a poor choice for exact tasks with one correct answer, like tax arithmetic — use a calculator or a database.
- Because behaviour comes from data, the same prompt can give slightly different answers each time.
A useful mental model
Think of a language model as an extremely well-read assistant with no memory of your business, no access to today's news unless you give it, and a strong instinct to always produce something plausible. Your job as the operator is to supply context and to check the output.