String Seed of Thought: Prompting LLMs for Distribution-Faithful and Diverse Generation

SSoT: A simple prompting method that enables LLMs to generate distribution-faithful and diverse outputs by first generating a random string.

This page requires Javascript. Please enable it to view the website. tl;dr We introduce String Seed of Thought (SSoT), a simple prompting method that substantially improves LLMs' ability to follow probabilistic instructions and generate diverse outputs. SSoT instructs the LLM to first generate a random string, then manipulate it to derive the answer. The method requires no additional training or external tools, only a change to the prompt. SSoT instructs the LLM to (1) generate a random string to create a seed, then (2) manipulate it to generate an output. This simple two-stage prompt…

saved by

• Colleen Chang

related reading

• Defeating Nondeterminism in LLM Inference - Thinking Machines Labthinkingmachines.ai
• [2201.11903] Chain of Thought Prompting Elicits Reasoning in Large Language Modelsarxiv.org
• Stream of Search (SoS): Learning to Search in Languagearxiv.org
• Modifying LLM Beliefs with Synthetic Document Finetuningalignment.anthropic.com
• Prompt Injection as Role Confusionrole-confusion.github.io
• Spilling the Beans: Teaching LLMs to Self-Report Their Hidden Objectivesarxiv.org
• 2402.07927arxiv.org
• Chain-of-Thought Promptinglearnprompting.org
• Chain-of-Thought Prompting | Prompt Engineering Guidepromptingguide.ai
• Self-Consistency | Prompt Engineering Guidepromptingguide.ai
• Towards a Typology of Strange LLM Chains-of-Thought1a3orn.com
• Prompt Engineering | Lil'Loglilianweng.github.io