System Prompts vs User Prompts and Why the Distinction Matters

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-ai-foundations-system-prompts-final1-creators

1. What is the system prompt's role?

   1. Just another user message
   2. The model's standing orders, more privileged and persistent than user messages
   3. The output formatter
   4. A network setting

2. Which is a strong use of the system prompt?

   1. One-off jokes
   2. User-specific data
   3. Establishing a consistent persona, format, or constraint across a conversation
   4. Secrets

3. What can the system prompt pin that's hard to override?

   1. Hardware settings
   2. Network speed
   3. File permissions
   4. Rules the user cannot easily flip mid-conversation

4. Why set output format in the system prompt?

   1. So you don't repeat the format instruction every turn
   2. Because user prompts can't set format
   3. To increase cost
   4. To slow the model

5. Where should refusal rules be defined?

   1. In every user prompt
   2. In the system prompt as part of standing orders
   3. In the README only
   4. Nowhere

6. Are system prompts unbypassable?

   1. Yes, always unbypassable
   2. Only on Tuesday
   3. No — clever users can sometimes override them
   4. Only in JSON

7. Can every rule be encoded perfectly in 100 words?

   1. Yes, always
   2. Only in haiku
   3. Never
   4. No — system prompts have limits; some behaviors need fine-tuning

8. Why don't secrets and access controls belong in system prompts?

   1. Users can extract them and replay/log layers may store them
   2. Secrets break JSON
   3. Secrets cost more
   4. Secrets disable the model

9. Which experiment shows the system prompt's persistence?

   1. Run one user message and stop
   2. Set 'You are a JSON-only API' in system, then send several user messages and see format hold
   3. Set system to empty
   4. Send only an emoji

10. What is a 'persona' in this context?

    1. A user account
    2. A login
    3. A consistent voice or role the model adopts across turns
    4. A network address

11. Why does role hierarchy matter?

    1. All messages are equal
    2. Hierarchy is decorative
    3. Hierarchy doubles cost
    4. Higher-privilege messages constrain lower-privilege ones

12. Which is a misuse of the system prompt?

    1. Pasting API keys or PII into it
    2. Setting an output format
    3. Setting a persona
    4. Defining refusal rules

13. Why is fine-tuning sometimes preferable to a system prompt?

    1. System prompts are always better
    2. Deeply ingrained behaviors are more reliable when trained in
    3. Fine-tuning is faster than prompts
    4. Fine-tuning is free

14. What does treating the system prompt like 'just another prompt' waste?

    1. Bandwidth
    2. Disk space
    3. Its main feature: persistent, privileged instruction
    4. CPU

15. Which is a sound default for new product builders?

    1. Set everything in user
    2. Skip the system prompt
    3. Encode secrets in system
    4. Set persona, format, and refusal rules in system; vary task in user