Designing Streaming UX That Survives Model Errors

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-tools-streaming-ux-r12a1-creators

1. What is a primary failure mode that streaming introduces in AI user interfaces?

   1. The interface automatically retries failed requests without user consent
   2. Connections may drop mid-token, leaving partial output visible
   3. Responses become too fast for users to read meaningfully
   4. Streaming causes all users to see identical responses

2. For chat-style AI responses, what is the recommended token rendering strategy?

   1. Render tokens incrementally as they arrive from the model
   2. Hide all streaming and show a loading spinner until complete
   3. Only display tokens after running a validation check on each one
   4. Buffer all tokens until the complete message arrives before displaying

3. When streaming structured output like JSON, what should the UI do when tokens arrive?

   1. Show a skeleton loader until the entire response completes
   2. Display each token immediately to show real-time progress
   3. Stream the partial JSON but mark invalid sections as errors
   4. Buffer tokens until the JSON is parseable, then update the UI

4. What should the UI display when a stream ends without a proper terminator?

   1. The partial content as-is, letting users copy what exists
   2. A clear error state indicating the stream was incomplete
   3. An automatic retry button that resubmits the request
   4. Nothing - the interface should remain silent about failures

5. What can an AI model do when an upstream connection is cancelled mid-token?

   1. Nothing - it cannot recover a cancelled stream
   2. Buffer the partial tokens and wait for reconnection
   3. Recover the stream and complete the response
   4. Automatically reconnect and resume from where it left off

6. Why should you avoid treating partial JSON as if it were complete?

   1. Partial JSON consumes too much memory to display
   2. JSON is too complex to render partially anyway
   3. The parser will crash if given incomplete JSON structures
   4. Partial JSON is invalid and would cause errors if processed

7. When persisting streamed AI output to a database, what is the safe approach?

   1. Only commit to the database when the stream completes
   2. Use a separate table for streaming and merge later
   3. Save partial data but mark it as 'incomplete' in a status field
   4. Commit each token as it arrives for real-time saving

8. What does SSE stand for in the context of streaming AI responses?

   1. Synchronous Stream Endpoint
   2. Secure Socket Encryption
   3. Server-Sent Events
   4. Structured Stream Exchange

9. A developer wants to show users that streaming is working. For prose output, what provides the best user experience?

   1. A progress percentage showing how much is left
   2. A preview of the first sentence before continuing
   3. A spinning loader that appears before content starts
   4. Incremental token display as the model generates them

10. What happens if you display streaming tokens for JSON before validation?

    1. The display automatically corrects syntax errors
    2. Users see malformed content that may confuse their app
    3. Users get a performance boost from seeing earlier content
    4. Nothing different - validation happens server-side anyway

11. Why is it important to handle streaming errors differently than regular request errors?

    1. The stream may partially succeed before failing, requiring special UI handling
    2. Streaming errors are the same as regular errors
    3. Streaming errors are faster and need quicker responses
    4. Regular errors don't affect user experience as much

12. A product shows partial JSON to users while waiting for the stream to complete. What's the problem with this approach?

    1. The approach is actually recommended for better UX
    2. JSON cannot be displayed in real-time due to size
    3. Users might copy and use invalid JSON that will break their code
    4. It uses too much bandwidth to display partial data

13. What is the main advantage of streaming tokens for chat interfaces?

    1. It provides immediate feedback while the model generates
    2. It guarantees the response will be completely accurate
    3. It reduces the computational cost of the model
    4. It makes the AI respond faster than non-streaming

14. A user experiences a connection drop during streaming. What should the interface do?

    1. Wait silently for the connection to restore itself
    2. Show the partial content with an error indicator
    3. Display a success message since streaming started
    4. Automatically retry the request in the background

15. What is a consequence of committing partial streamed output to a database?

    1. The model learns from the partial data incorrectly
    2. You store half-rendered messages that cause problems later
    3. Partial commits improve database performance
    4. The database automatically fixes incomplete records