Research map / Large language models / Reasoning and chain of thought
STaR: Bootstrapping Reasoning With Reasoning
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 |
|---|---|---|---|
| add rationalization (answer-hinted rationale generation for failed problems) (ablation) | STaR · CommonsenseQA dev | accuracy: 68.8 -> 72.5 | Helped |
| add rationalization (ablation) | STaR · GSM8K test | accuracy: 10.1 -> 10.7 | No clear effect |
| STaR rationale bootstrapping vs direct-answer finetuning | GPT-J · CommonsenseQA dev | accuracy: 60.0 -> 72.5 | Helped |
| STaR vs direct-answer finetuning | GPT-J · GSM8K test | accuracy: 5.8 -> 10.7 | Helped |
| STaR vs direct finetuning without rationales | GPT-J · Arithmetic | accuracy: 76.3% -> 89.5% | Helped |
| include few-shot prompts during fine-tuning (ablation) | STaR without rationalization · CommonsenseQA dev | accuracy: 60.9% -> 68.8% | Helped |
| include few-shot prompts during fine-tuning (ablation) | STaR with rationalization · CommonsenseQA dev | accuracy: 69.9% -> 72.5% | Helped |
| use higher sampling temperature (0.5 or 0.7) instead of rationalization (ablation) | STaR · Arithmetic / CQA | accuracy | Hurt |
Closest papers by meaning
- Tailoring Self-Rationalizers with Multi-Reward Distillation (ICLR 2024 · similarity 0.89)
- BRiTE: Bootstrapping Reinforced Thinking Process to Enhance Language Model Reasoning (ICML 2025 · similarity 0.88)
- General-Reasoner: Advancing LLM Reasoning Across All Domains (NeurIPS 2025 · similarity 0.88)
- Decompose, Analyze and Rethink: Solving Intricate Problems with Human-like Reasoning Cycle (NeurIPS 2024 · similarity 0.88)
- SuperCorrect: Advancing Small LLM Reasoning with Thought Template Distillation and Self-Correction (ICLR 2025 · similarity 0.88)
- Improving Factuality and Reasoning in Language Models through Multiagent Debate (ICML 2024 · similarity 0.88)
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.