AI On-Device Models: Phi, Gemma, and the Edge Tradeoff

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-model-families-on-device-models-final5-creators

1. What is a primary advantage of running AI models directly on personal devices like phones and laptops?

   1. Requires constant internet connection
   2. Higher reasoning quality than cloud models
   3. Privacy-preserving local inference
   4. Ability to handle unlimited context lengths

2. Which of the following describes a key limitation of on-device AI models compared to cloud-based flagship models?

   1. Cannot match flagship reasoning quality
   2. Cannot operate without internet connectivity
   3. Require payment for each inference
   4. Are always slower than cloud models

3. A developer is building an app that processes sensitive financial information on a user's device. Which approach aligns with the lesson's guidance?

   1. Route the workload to an on-device model to preserve privacy
   2. Avoid AI entirely for financial applications
   3. Always send sensitive data to cloud for better accuracy
   4. Use cloud inference without informing the user

4. What does the lesson recommend when an on-device model cannot handle a complex query?

   1. Escalate to cloud with user consent
   2. Automatically send data to cloud without notification
   3. Refuse to answer the query
   4. Switch to a larger on-device model regardless of memory

5. Which statement about on-device AI model performance is correct?

   1. On-device models perform best on the latest flagship devices
   2. Performance is identical on all devices running the same model
   3. Older devices always run on-device models faster than cloud
   4. Performance varies significantly across different hardware devices

6. What cost advantage do on-device AI models offer after initial deployment?

   1. Free but with limited usage caps
   2. Zero cost per inference after deployment
   3. One-time licensing fee only
   4. Lower cost per inference than cloud APIs

7. Which scenario would be least appropriate for on-device AI deployment?

   1. Complex multi-step reasoning across a long document
   2. Local photo organization with AI tags
   3. Private note-taking with autocomplete
   4. Offline language translation app

8. What latency characteristic makes on-device AI attractive for certain applications?

   1. Latency that remains constant regardless of device age
   2. Guaranteed faster latency than all cloud models
   3. Latency that improves with internet speed
   4. Predictable latency without network dependency

9. When testing on-device AI performance, what device should developers primarily test against?

   1. A device with maximum RAM
   2. The most expensive device available
   3. Their bottom-quartile target device
   4. The latest flagship development phone

10. Which term describes AI processing that occurs directly on a user's device rather than on remote servers?

    1. Edge inference
    2. Distributed networking
    3. Server-side processing
    4. Cloud computing

11. What happens to on-device AI performance when attempting to process very long input contexts?

    1. Performance improves automatically
    2. Context length has no impact
    3. Memory costs increase significantly
    4. Only cloud models can process any length

12. Which of these is NOT listed as an advantage of on-device AI models in the lesson?

    1. Zero cost per inference after deployment
    2. Privacy-preserving local inference
    3. Predictable latency without network
    4. Superior reasoning compared to cloud models

13. What is the primary capability gap between on-device models and cloud flagship models?

    1. Ability to run on phones
    2. Support for multiple languages
    3. Reasoning quality and complexity handling
    4. Energy efficiency

14. A user needs AI assistance while on an airplane with no WiFi. Which capability makes on-device models suitable for this scenario?

    1. Require less battery than cloud models
    2. Are only usable with fast internet
    3. Can operate without network connectivity
    4. Automatically connect to available networks

15. Which example represents an inappropriate use case for on-device AI based on the lesson content?

    1. On-device code completion for simple functions
    2. Private calendar event extraction
    3. Local voice memo transcription
    4. Real-time video analysis requiring flagship-quality object detection