Research map / Large language models / Language, safety and interpretability
String Seed of Thought: Prompting LLMs for Distribution-Faithful and Diverse Generation
Kou Misaki, Takuya Akiba
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 the String Seed of Thought instruction (generate a random string, manipulate it to sample) | deepseek-r1-0528 · PIF biased 2-choice | JS divergence (x1e-3): 69.58 -> 1.51 | Improved |
| add SSoT to gpt-4o | gpt-4o-2024-08-06 · PIF biased 3-choice | JS divergence (x1e-3): 113.06 -> 9.75 | Improved |
| add SSoT to QwQ-32B on unbiased 2-choice | QwQ-32B · PIF 2-choice | JS divergence (x1e-3): 2.43 -> 3.39 | Hurt |
| add SSoT to deepseek-v3 on biased 9-choice | deepseek-v3-0324 · PIF biased 9-choice | JS divergence (x1e-3): 297.33 -> 44.90 | Improved |
| use prompt ensemble instead of SSoT (ablation) | deepseek-r1 PIF · PIF varying action spaces | JS divergence | Mixed |
| supply an external randomized seed string instead of LLM-generated string (ablation) | deepseek-r1 PIF · PIF varying action spaces | JS divergence | Improved |
| SSoT prompt in adversarial Rock-Paper-Scissors | LLM player vs 10 Black Belt bots · RPS Dojo, 100 games per match | score | Improved |
| add SSoT to open-ended generation | deepseek-r1 (T=0.6) · NoveltyBench curated | Distinct: 4.70 -> 6.19 | Improved |
| add SSoT to WildChat prompts | deepseek-r1 (T=0.6) · NoveltyBench WildChat | Distinct (Utility): 3.39 (4.08) -> 5.25 (4.86) | Improved |
| raise temperature to 1.0 instead of using SSoT (ablation) | deepseek-r1 · NoveltyBench curated | Distinct: 4.70 -> 5.57 | Improved |
| lengthen reasoning (more thinking tokens) (ablation) | s1.1-32B random integer generation · integers 0-127 | uniformity (KL/JS) | Improved |
| lengthen reasoning (more thinking tokens) (ablation) | deepseek-r1 style generation of random strings at T=0 · LZ complexity and zlib compression | string complexity | Improved |
Closest papers by meaning
- Evolving Prompts In-Context: An Open-ended, Self-replicating Perspective (ICML 2025 · similarity 0.89)
- StrategyLLM: Large Language Models as Strategy Generators, Executors, Optimizers, and Evaluators for Problem Solving (NeurIPS 2024 · similarity 0.88)
- Scaling Prompt Synthesis for Large Language Model Reasoning (ICML 2026 · similarity 0.88)
- From Noise to Diversity: Random Embedding Injection in LLM Reasoning (NeurIPS 2026 · similarity 0.87)
- Reward Is Enough: LLMs Are In-Context Reinforcement Learners (ICLR 2026 · similarity 0.87)
- Boosting of Thoughts: Trial-and-Error Problem Solving with Large Language Models (ICLR 2024 · similarity 0.87)
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.
