AI Paper Atlas

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

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Statistical frameworks for LLM evaluation

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 changedWhereMeasuredEffect
add the String Seed of Thought instruction (generate a random string, manipulate it to sample)deepseek-r1-0528 · PIF biased 2-choiceJS divergence (x1e-3): 69.58 -> 1.51Improved
add SSoT to gpt-4ogpt-4o-2024-08-06 · PIF biased 3-choiceJS divergence (x1e-3): 113.06 -> 9.75Improved
add SSoT to QwQ-32B on unbiased 2-choiceQwQ-32B · PIF 2-choiceJS divergence (x1e-3): 2.43 -> 3.39Hurt
add SSoT to deepseek-v3 on biased 9-choicedeepseek-v3-0324 · PIF biased 9-choiceJS divergence (x1e-3): 297.33 -> 44.90Improved
use prompt ensemble instead of SSoT (ablation)deepseek-r1 PIF · PIF varying action spacesJS divergenceMixed
supply an external randomized seed string instead of LLM-generated string (ablation)deepseek-r1 PIF · PIF varying action spacesJS divergenceImproved
SSoT prompt in adversarial Rock-Paper-ScissorsLLM player vs 10 Black Belt bots · RPS Dojo, 100 games per matchscoreImproved
add SSoT to open-ended generationdeepseek-r1 (T=0.6) · NoveltyBench curatedDistinct: 4.70 -> 6.19Improved
add SSoT to WildChat promptsdeepseek-r1 (T=0.6) · NoveltyBench WildChatDistinct (Utility): 3.39 (4.08) -> 5.25 (4.86)Improved
raise temperature to 1.0 instead of using SSoT (ablation)deepseek-r1 · NoveltyBench curatedDistinct: 4.70 -> 5.57Improved
lengthen reasoning (more thinking tokens) (ablation)s1.1-32B random integer generation · integers 0-127uniformity (KL/JS)Improved
lengthen reasoning (more thinking tokens) (ablation)deepseek-r1 style generation of random strings at T=0 · LZ complexity and zlib compressionstring complexityImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy