Research map / Large language models / Agents, code and math
Decomposed Prompting: A Modular Approach for Solving Complex Tasks
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 regex answer extraction after CoT with a GPT3 answer-extraction sub-task handler | CoT math reasoning · MultiArith | accuracy: +17 pts (78 to 95) | Helped |
| replace regex answer extraction after CoT with a GPT3 answer-extraction sub-task handler | CoT math reasoning · GSM8K | accuracy: +14 pts (36 to 50.6) | Helped |
| decompose letter concatenation into split, str_pos, merge sub-tasks instead of a single CoT | k-th letter concatenation · Letter concatenation (k=3, space delimiter, longer inputs) | exact match | Helped |
| recursive decomposition of list reversal (split halves) vs base CoT | List reversal with davinci-001 · Reversal, 4-10 items | exact match | Helped |
| unrolling the decomposition into a single CoT (rolled-out) for reversal (ablation) | List reversal · Reversal, longer lists | exact match | Hurt |
| fine-grained vs coarse-grained decomposition (more question-type sub-tasks) on CommaQA (ablation) | CommaQA-E · CommaQA-E | exact match | Helped |
| decomposition into sub-task handlers (vs CoT) on CommaQA-E | CommaQA-E · CommaQA-E (coarse and fine decomposition) | exact match | Helped |
| retrieval via Elasticsearch-based retrieve_odqa decomposition vs NoDecomp retrieval baseline (Flan-T5-XXL) | Open-domain multihop QA · 2WikiMultihopQA, MuSiQue, HotpotQA | answer F1 | Helped |
| Elasticsearch-based decomposed retrieval with Codex vs NoDecomp retrieval on HotpotQA | Open-domain multihop QA (Codex) · HotpotQA | answer F1 | No clear effect |
| decomposition with differently-structured sub-question generation (GPT3 generates each question instead of looping) for letter concatenation (ablation) | k-th letter concatenation · Letter concatenation | exact match | No clear effect |
| tail-recursive reversal (reverse(list[1:]) + head) instead of halving (ablation) | List reversal · Reversal, longer inputs | exact match | Mixed |
Closest papers by meaning
- Scaling Prompt Synthesis for Large Language Model Reasoning (ICML 2026 · similarity 0.89)
- In-Context Principle Learning from Mistakes (ICML 2024 · similarity 0.89)
- Improving Complex Reasoning with Dynamic Prompt Corruption: A Soft Prompt Optimization Approach (ICLR 2025 · similarity 0.88)
- Large Language Models as Analogical Reasoners (ICLR 2024 · similarity 0.88)
- StrategyLLM: Large Language Models as Strategy Generators, Executors, Optimizers, and Evaluators for Problem Solving (NeurIPS 2024 · similarity 0.88)
- GReaTer: Gradients Over Reasoning Makes Smaller Language Models Strong Prompt Optimizers (ICLR 2025 · 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.