Building A Custom GPT For A Specific Workflow

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-openai-custom-gpts-creators

1. According to the narrow-scope principle, which Custom GPT is most likely to produce reliable, consistent results?

   1. A Custom GPT that answers customer service questions
   2. A Custom GPT that serves as a general marketing assistant
   3. A Custom GPT that helps with any writing task
   4. A Custom GPT that drafts LinkedIn posts from Loom video transcripts in your company's brand voice

2. What security concern exists when uploading documents to a publicly-shared Custom GPT?

   1. OpenAI employees will review and potentially share the uploaded documents
   2. The knowledge files will eventually get deleted by the platform
   3. The files may be automatically indexed and appear in search engine results
   4. Motivated users can potentially extract knowledge-file contents through prompt injection techniques

3. What is the recommended structure for an effective system prompt in a Custom GPT?

   1. ROLE, INPUT, OUTPUT, RULES, NEVER, and EDGE CASES sections
   2. Background, context, personality, tone, and examples
   3. System message, user message, assistant message, and conversation history
   4. Task description, constraints, output format, and warnings

4. What should you do before sharing a Custom GPT with others, according to the applied exercise in the lesson?

   1. Publish the GPT immediately to get user feedback faster
   2. Ask friends to review the prompt for grammatical errors
   3. Test it once with a sample input to verify it works
   4. Run the GPT on at least 10 real inputs, log failures, and tighten the prompt based on results

5. A student builds a Custom GPT called 'Legal Helper' that can review any type of legal document from any jurisdiction. Why is this likely to fail?

   1. Legal documents require fine-tuning rather than a system prompt
   2. The Custom GPT builder doesn't support legal documents
   3. The scope is too broad — it lacks a specific input shape and output constraint
   4. The knowledge files would be too small to be useful

6. What does the lesson identify as the 'hard part' of building a useful Custom GPT?

   1. Uploading knowledge files to the builder interface
   2. Selecting the right model version from OpenAI's catalog
   3. Writing a system prompt narrow enough to do one job well
   4. Configuring HTTP actions to call external APIs

7. A builder uploads a 500-page employee handbook as a knowledge file to their company's Custom GPT. What does the lesson suggest about this approach?

   1. It's necessary to ensure the GPT understands the full context
   2. It's acceptable if the document is converted to a specific file format
   3. It's ideal because the GPT can access all company information in one place
   4. It's counterproductive — large documents often hurt more than help due to inconsistent referencing

8. What does the lesson mean by 'scoping' a Custom GPT?

   1. Defining exactly what input the GPT will receive, what output it will produce, and what it should never do
   2. Deciding how much to charge users for access to the GPT
   3. Choosing which AI model to use (GPT-4 vs GPT-3.5)
   4. Selecting the color scheme and name for the GPT's launch tile

9. Based on the lesson, what is an 'action' in the context of Custom GPTs?

   1. An HTTP API call the GPT can make to external services during a conversation
   2. A knowledge file that provides reference material
   3. A safety filter that prevents harmful outputs
   4. A rule in the system prompt that constrains the GPT's behavior

10. The lesson mentions that community members have observed an arc where the first Custom GPT is usually disappointing. What is the recommended solution?

    1. Give up after the first failure and try a different platform
    2. Use more knowledge files to make the first GPT more capable
    3. Hire a professional prompt engineer to write the first prompt
    4. Keep building and iterating — the third or fourth GPT tends to be narrow and useful

11. Why does the lesson recommend versioning knowledge files directly in the system prompt?

    1. To prevent users from downloading the knowledge files
    2. To help users understand when the file was last modified
    3. To ensure the GPT references the correct version and users get consistent results
    4. To comply with OpenAI's terms of service for Custom GPTs

12. A builder wants to create a Custom GPT that accesses their company's internal pricing database. What does the lesson advise about this use case?

    1. Internal data is safe to share in any Custom GPT
    2. Pricing databases should be uploaded as knowledge files
    3. Use a private workspace GPT instead of a public Custom GPT for sensitive data
    4. This is an ideal use case for a public Custom GPT

13. What is the core insight of the statement 'a great Custom GPT does one job, a bad Custom GPT tries to be a coworker'?

    1. Coworkers are not good at providing feedback on GPT outputs
    2. One job means less than five tasks per day
    3. Custom GPTs should aim to replace human coworkers entirely
    4. Narrow, specific tasks produce better results than broad, assistant-like behavior

14. What happens when a Custom GPT receives an input that doesn't match its defined input shape, according to the system prompt skeleton?

    1. The GPT should ask the user to rephrase their input indefinitely
    2. The GPT should follow an explicit fallback defined in the EDGE CASES section
    3. The GPT should expand its capabilities to handle the new input
    4. The GPT should provide a generic error message and shut down

15. Based on the lesson, what is the best way to identify if your task is too broad for a Custom GPT?

    1. Ask five friends if they think the task is a good idea
    2. Try to describe the input and output in one sentence each — if you can't, the task is too broad
    3. Search Reddit to see if anyone has built something similar
    4. Test the task with 100 different inputs before building