AI 101

Lesson 2 · about 6 min

What it's good and bad at

Knowing where these tools are strong and where they fail is what keeps you from trusting a confident wrong answer. The short version: a language model is very good with words and shaky with anything that has one correct answer.

Strong: shaping and explaining language

The model’s whole skill is producing fluent text. So it is genuinely useful when the job is about words:

For example: paste in a rude email from your landlord and ask for a calm reply that still makes your point. You get a solid first draft in seconds. You can read it and fix it yourself, because you already know what you meant. That is the sweet spot — you stay the editor.

Weak: exact facts, math, and recent events

The same machine struggles when there is a single right answer it has to get exactly:

Hallucination: confidently wrong

When the model does not have a fact, it does not stop or say “I don’t know.” It predicts the most natural-sounding text and keeps going. The result is a hallucination — a made-up detail, delivered in the same calm, sure tone as a real one.

Here is a concrete case. You ask for three books on a niche topic. It gives you three titles, each with a real-sounding author and publisher. One of the books does not exist. Nothing in the wording warns you. The fake entry reads just like the real ones.

This happens because fluent text is the goal, not true text. And the model does not know what it does not know — there is no inner voice checking its own claims.

So the rule of thumb: use it for language, and check any hard fact yourself. Next, you will see how the way you ask changes what you get back.

Knowledge check

  1. Which task is a language model best suited for?
  2. You ask a model about an event from last week and it answers in confident detail. What is the safe assumption?
  3. "Hallucination," in the context of these tools, means:
  4. What is the best habit when an answer includes a specific fact, name, or number?