Least-to-Most Prompting Quiz

Answers and explanations appear as you take the quiz.

1. What are the two main stages of Least-to-Most prompting?
   • Decompose the problem into subquestions, then solve them in order
   • Train then fine-tune
   • Encode then decode
   • Sample many answers then vote
2. How does Least-to-Most differ from standard chain-of-thought prompting?
   • It uses no examples at all
   • It explicitly decomposes the problem and feeds each subanswer into the next step
   • It requires fine-tuning the model
   • It only works on images
3. What kind of task does the SCAN benchmark use to test compositional generalization?
   • Classifying satellite images of fishing vessels
   • Predicting fuel use from ship weather data
   • Interpreting natural-language commands that combine primitive actions
   • Aligning speech frames with handwritten characters
4. Why is the order of subquestions important?
   • It is not; order is random
   • Later questions must come first
   • Order only affects formatting
   • Each subquestion should only depend on ones already solved before it
5. How are intermediate answers used during the solving stage?
   • They are appended to the context so later steps can build on them
   • They are discarded after each step
   • They are sent to a separate model
   • They are averaged into one number
6. What is a key risk in a Least-to-Most chain?
   • It cannot use examples
   • A wrong early subanswer can corrupt all downstream steps
   • It only works in one language
   • It always requires a GPU cluster
7. What accuracy did the paper report for code-davinci-002 on the SCAN length split using Least-to-Most with 14 exemplars?
   • About 16%
   • Exactly 50%
   • At least 99% on the reported setup
   • Zero, because no demonstrations were used
8. Besides SCAN, which task areas did the Least-to-Most paper also evaluate?
   • Wildlife re-identification and acoustic classification
   • Ship routing and port scheduling
   • Image outpainting and watermark detection
   • Symbolic manipulation and math reasoning