Research map / Large language models / Reasoning and chain of thought
Self-Consistency Improves Chain of Thought Reasoning in Language Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace greedy CoT decoding with 40 sampled reasoning paths and answer majority voting | PaLM-540B · GSM8K | accuracy: 56.5 -> 74.4 (+17.9) | Helped |
| Replace greedy CoT decoding with sampled paths and answer aggregation | PaLM-540B · AQuA | accuracy: 35.8 -> 48.3 (+12.5) | Helped |
| Replace greedy CoT decoding with sampled paths and answer aggregation | PaLM-540B · StrategyQA | accuracy: 75.3 -> 81.6 (+6.3) | Helped |
| Replace greedy CoT decoding with sampled paths and answer aggregation | PaLM-540B · ARC-challenge | accuracy: 85.2 -> 88.7 (+3.5) | Helped |
| Use unweighted majority voting instead of normalized weighted answer summation (ablation) | Self-consistency · GSM8K | accuracy: 74.1 -> 74.4 | No clear effect |
| Use unnormalized rather than normalized sequence probabilities when weighting answers (ablation) | Self-consistency · GSM8K | accuracy: 74.1 -> 59.9 | Hurt |
| Use weighted average rather than weighted sum for answer aggregation (ablation) | Self-consistency · GSM8K | accuracy: 74.1 -> 22.1 | Hurt |
| Use beam search instead of sampling to generate paths (ablation) | Self-consistency on UL2-20B · AQuA | accuracy: 26.9 -> 24.2 | Hurt |
| Use beam search instead of sampling to generate paths (ablation) | Self-consistency on UL2-20B · MultiArith | accuracy: 14.7 -> 10.8 | Hurt |
| Add self-consistency over 40 paths to an imperfect CoT prompt (ablation) | LaMDA-137B · GSM8K | accuracy: 14.9 -> 23.4 | Helped |
| Add self-consistency to equation-based CoT prompting (ablation) | LaMDA-137B · GSM8K | accuracy: 5.0 -> 6.5 | Helped |
| Add self-consistency to zero-shot CoT (ablation) | PaLM-540B · GSM8K | accuracy: 43.0 -> 69.2 | Helped |
Closest papers by meaning
- Escape Sky-high Cost: Early-stopping Self-Consistency for Multi-step Reasoning (ICLR 2024 · similarity 0.92)
- Optimal Self-Consistency for Efficient Reasoning with Large Language Models (ICML 2026 · similarity 0.91)
- SELF-DISCOVER: Large Language Models Self-Compose Reasoning Structures (NeurIPS 2024 · similarity 0.91)
- Chain-of-Thought Reasoning Without Prompting (NeurIPS 2024 · similarity 0.91)
- How Does Chain of Thought Think? Mechanistic Interpretability of Chain-of-Thought Reasoning with Sparse Autoencoding (AAAI 2026 · similarity 0.91)
- Reliable Chain-of-Thought via Prefix Consistency (NeurIPS 2026 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.