Skip to content
ml-recall

#test-time-compute

2 questions from Alignment & Reasoning

Timed test

2 questions · 2 min · score and explanations at the end

Take a test
Q 1 / 2Score 0 / 0

Question 1: Short answer

easyAI-checkedAlignment & Reasoning

A model is prompted with chain-of-thought exemplars for a math word problem. Greedy decoding gives the answer 26. With self-consistency, six reasoning paths are sampled instead:

Path Final answer Length-normalized probability of the path
1 18 0.44
2 26 0.52
3 18 0.40
4 14 0.47
5 18 0.43
6 26 0.49

Aggregating by majority vote over the final answers, which answer does self-consistency return?

a number