2307.05300.pdf

– Large language models (LLMs) face challenges in knowledge-intensive and reasoning-intensive tasks.
– Current LLMs lack cognitive synergy and slow-thinking capabilities.
– Previous works have enhanced reasoning abilities but struggle with factual hallucination.
– Solo Performance Prompting (SPP) unleashes cognitive synergy in LLMs through multi-persona self-collaboration.
– SPP improves problem-solving abilities and reduces factual hallucination.
– SPP is evaluated on Trivia Creative Writing, Codenames Collaborative, and Logic Grid Puzzle tasks.
– Cognitive synergy only emerges in GPT-4 and not in less capable models.

– SPP is evaluated on GPT-4, GPT-3.5-turbo, and Llama2-13b-chat models.
– Cognitive synergy only emerges in GPT-4 and not in other models.

– Solo Performance Prompting (SPP) enhances problem-solving abilities in complex tasks.
– SPP reduces factual hallucination and maintains strong reasoning capabilities.
– Cognitive synergy emerges in GPT-4 but not in less capable models.

– The paper proposes Solo Performance Prompting (SPP) to enhance problem-solving in large language models.
– SPP engages in multi-turn self-collaboration with multiple personas to unleash cognitive synergy.
– Multiple fine-grained personas in LLMs improve problem-solving abilities compared to a single persona.
– SPP reduces factual hallucination and maintains strong reasoning capabilities in LLMs.
– Cognitive synergy only emerges in GPT-4 and not in less capable models.

– Solo Performance Prompting (SPP) transforms a single LLM into a cognitive synergist.
– SPP improves problem-solving abilities compared to using a single persona.
– SPP reduces factual hallucination and maintains strong reasoning capabilities.
– Cognitive synergy only emerges in GPT-4 and not in less capable models.

– Full results of the three tasks: Trivia Creative Writing, Codenames Collaborative, and Logic Grid Puzzle can be found in Tables 5, 6, and 7, respectively.

– Solo Performance Prompting (SPP) is a method to make a computer smarter.
– It uses different “personas” to help the computer solve problems.
– SPP improves problem-solving abilities compared to other methods.
– It reduces mistakes and maintains strong reasoning capabilities.
– SPP works best in advanced models like GPT-4.