Free 🧰 Practical LLM Lessons

LLM Lab 5: Instructions and Guardrails

See how system messages, user requests, and safety policies stack to shape the final reply.

Published January 8, 2026

Previous

LLM Lab 4: Creative vs Precise

Next

LLM Lab 6: Chain-of-Thought Builder

Who is really in charge?

LLMs weigh multiple instruction layers: system, developer, and user. Later messages can be ignored if they conflict with higher-priority rules.

Can You Guess?

A chat has: System: 'Be a concise assistant.' User: 'Write a 1,000-word essay.' Which wins?

Instruction priority stack

Follow the steps and watch each checkpoint light up as you progress.

1

Step 1: System

Highest priority: sets behavior and safety guardrails.

2

Step 2: Developer

Optional middle layer: templates or app-specific rules.

3

Step 3: User

Lowest priority: the prompt you type.

4

Step 4: Policies

Safety checks can refuse or redact content even if prompted otherwise.

Try this now

Flip the instructions deliberately

  • Send a system prompt: “You are a strict 100-word limit assistant.”
  • Then ask for something long. See how the model enforces the cap.
  • Remove the system message and compare the response length.

    • Key Takeaways

    • ✓ Instruction order and type set priorities.
    • ✓ Conflicts create unpredictable responses.
    • ✓ Clear, non-conflicting instructions yield stable outputs.

    Quick poll

    What brought you to this free lesson?

    Lesson: LLM Lab 5: Instructions and Guardrails

    ← Previous Lesson LLM Lab 4: Creative vs Precise Next Lesson → LLM Lab 6: Chain-of-Thought Builder
    Back to Library Unlock More Lessons