Temperature Tuning and Sampling: Determinism by Task

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-prompting-LLM-temperature-tuning-by-task-creators

1. What does setting temperature to 0 actually do in an LLM?

   1. It disables the neural network entirely
   2. It forces the model to always select the most probable next token at each step
   3. It increases the model's accuracy
   4. It makes the model completely random

2. A developer wants completely reproducible outputs from an LLM for testing purposes. What should they do in addition to setting temperature to 0?

   1. Add more examples to the prompt
   2. Switch to a different API provider
   3. Pin the random seed if the provider exposes this option
   4. Use a larger model

3. Which task from the recommended temperature table uses a setting of 0.9?

   1. Classification
   2. Extraction
   3. Brainstorming
   4. Code generation

4. Why does the lesson recommend storing temperature values in a configuration table in code rather than hardcoding them directly in function calls?

   1. Tables are faster to execute than variables
   2. Different tasks require different temperatures, and a table makes adjustment easier without code changes
   3. The AI model enforces using tables
   4. Hardcoded values cause security vulnerabilities

5. A student writes a vague, unclear prompt but sets temperature to 0, expecting precise results. What does the lesson indicate about this approach?

   1. Temperature cannot make up for a bad prompt—the prompt itself must be clear
   2. Low temperature compensates for unclear prompts
   3. Temperature 0 makes prompts irrelevant
   4. The model will ask for clarification

6. You call an LLM API twice with identical parameters including temperature 0, but receive different outputs. What is the most likely explanation?

   1. Tie-breaking randomness still occurs even at temperature 0
   2. The API service is down
   3. The model has a bug
   4. You accidentally changed the prompt

7. What temperature range does the lesson recommend for drafting business prose like emails and reports?

   1. 0.3-0.5 (moderate creativity)
   2. 1.0+ (maximum randomness)
   3. 0.7-1.0 (high creativity)
   4. 0.0-0.1 (fully deterministic)

8. A company updates their LLM to a new model version but keeps the same temperature settings. What does the lesson recommend?

   1. Keep the same settings—they work fine
   2. Use the default temperature of the new model
   3. Always increase temperature after updates
   4. Re-tune temperature for the new model version

9. What is the primary reason temperature should be treated as a 'tested config' rather than a fixed value?

   1. Fixed values cause errors
   2. Optimal temperature varies by task, model version, and use case
   3. Tested configs are more secure
   4. Testing is optional but recommended

10. Which statement best captures what the lesson means by calling temperature a 'per-task parameter'?

    1. Temperature is determined by the hardware
    2. Temperature should be set differently depending on the type of task you're doing
    3. Each model has one fixed temperature
    4. All tasks should use the same temperature

11. What temperature setting does the lesson recommend for extraction tasks?

    1. 0.9
    2. 0.0
    3. 0.1
    4. 0.5

12. If you set temperature too high for a classification task, what is the most likely negative outcome?

    1. The model may produce inconsistent or incorrect labels
    2. The model will refuse to classify
    3. The classification will complete faster
    4. Classification accuracy will improve

13. The lesson describes temperature as 'not a vibe knob.' What does this metaphor mean?

    1. Temperature knobs are illegal
    2. Temperature only affects creative tasks
    3. Temperature has no effect on output
    4. Temperature should be set deliberately based on task requirements, not adjusted casually

14. What does the lesson say happens when behavior 'drifts' in an LLM application?

    1. You should revisit your temperature settings as part of troubleshooting
    2. You should ignore it
    3. You should lower the temperature
    4. You should switch models immediately

15. From the lesson's temperature table, what value is assigned to the 'summary' task?

    1. 0.5
    2. 0.1
    3. 0.3
    4. 0.0