2 · Reasoning & Logic

2. Chain-of-Thought & Self-Consistency Sampling

Step-by-step cognitive scaffolding, intermediate scratchpads, and majority vote decoding.

9 min read · 2 MCQs

Chain-of-Thought (CoT) Mechanics

Autoregressive LLMs compute output token-by-token. Asking the model to produce step-by-step reasoning tokens before stating the final answer creates an internal computational scratchpad that allows deeper inference passes on complex logic and arithmetic.

Self-Consistency Majority Voting

For mission-critical reasoning, generate multiple outputs at temperature $T=0.7$ with CoT enabled, extract the final answers across all paths, and select the answer via majority vote. This filters out stochastic reasoning errors.

Chapter quiz

2 questions · pass mark 75%

1. Why does Chain-of-Thought (CoT) prompting improve multi-step mathematical accuracy?
2. What is self-consistency sampling?