Llm Prompting

# LLM prompting

Prompting is the art and engineering of instructing a language model to produce the desired output. Although it's sometimes trivialized as "writing well", there are specific patterns, technical trade-offs, and architectural decisions behind the prompt that separate a reliable system from one that works by miracle.

## System prompt vs user prompt

Modern LLMs accept two main instruction channels:

- **System prompt**: persistent instructions that define role, behavior, constraints. Processed before any user message. The model treats them with higher priority than instructions injected into a user message.
- **User prompt**: what the end user writes. Treated as input to process.