Standard Practices for Text Generation

Part of the Prompt Engineering course

Lessons two and three gave you a feel for how the model actually produces text one token at a time, and the five principles for steering it. This lesson turns that understanding into a small set of habits that make text generation reliable enough to build on. These are the conventions practitioners reach for every day, and once they are second nature your prompts stop surprising you.

When your prompt contains both an instruction and a chunk of text to act on, the model has to figure out where one ends and the other begins. If you just run them together, it can misread part of your data as a new instruction, or fold your instruction into the thing it is supposed to be summarizing.