Self-Consistency Decoding Quiz

Answers and explanations appear as you take the quiz.

1. How does self-consistency decoding pick a final answer from a language model?
   • Sample many reasoning paths and take a majority vote on the answer
   • Always pick the single most likely token at each step
   • Train the model on its own outputs repeatedly
   • Remove all randomness from generation
2. Under what condition can self-consistency improve answer accuracy?
   • Whenever the number of samples increases
   • When correct sampled paths converge and errors are not too strongly correlated
   • When every reasoning chain has identical wording
   • Only when the model weights are retrained
3. What decoding method does self-consistency replace?
   • Backpropagation
   • Tokenization
   • Greedy decoding of a single chain
   • Weight quantization
4. Which part of each sampled chain is actually used in the final vote?
   • The full reasoning text
   • The number of tokens generated
   • The temperature setting
   • Only the final answer
5. What is the main cost of increasing the number of self-consistency samples?
   • More inference compute; accuracy gains depend on the task and can have diminishing returns
   • It retrains the model weights after every response
   • It guarantees that the majority is correct
   • It eliminates the need to sample any reasoning paths
6. Self-consistency is an early example of which broader concept?
   • Model compression
   • Inference-time scaling (spending extra compute to reason better)
   • Data augmentation during training
   • Hardware acceleration
7. Why may answer normalization be useful before a self-consistency vote?
   • It makes all reasoning paths independent
   • It verifies the answer against external evidence
   • It can group semantically equivalent answers that use different formatting
   • It changes the model’s sampled probabilities
8. What is a key limitation of a majority answer from self-consistency?
   • A majority vote cannot be computed for text
   • Sampling always changes the model weights
   • Every reasoning path is guaranteed independent
   • Correlated reasoning errors can make an incorrect answer win the vote