Introduction to Prompting › The Challenge of Communication

Here's something that trips up many beginners: language models don't "understand" in the way humans do. When you ask a question, the model isn't thinking about your intent or drawing on life experience. It's pattern-matching against billions of examples from its training data to predict what response would best fit your prompt.

This has practical implications. A human colleague might pick up on subtle hints or fill in missing context based on your shared history. A language model can't do that unless you explicitly provide the context. It doesn't remember previous conversations (unless you include that history in your prompt), and it doesn't know anything about you, your project, or your goals beyond what you tell it.

Once you understand this limitation, you can account for it in your prompts. Instead of expecting the model to read your mind, you learn to communicate in a way that uses its strengths. You provide the context it needs. You structure your requests clearly. You iterate when the first attempt doesn't quite work.

This is why prompting is both an art and a science. The science is understanding how models process input and generate output. The art is crafting prompts that guide the model toward your desired result. Both aspects improve with practice.