Literary techniques help LLMs simulate human-like personalities

[2310.07971] A Literary Character Approach to Human-like … – arXiv
(2023-10-12) Abstract: Large language models (LLMs) have demonstrated remarkable abilities in simulating human-like conversations. However, their potential in accurately simulating human personality traits for social science research remains underexplored. This study introduces a novel “literary character approach” to enhance LLMs’ persona simulation capabilities. By constructing detailed “character skeletons” based on established psychological frameworks and enriching them with nuanced, literary-style descriptions, we guide LLMs to generate responses that are more consistent with specific personality profiles. We evaluated our method by comparing the personality traits of LLM-generated texts with those of real individuals from a public dataset. The results indicate that our approach significantly improves the alignment of LLM-generated personas with target personality profiles, as measured by standard psychological metrics. This work lays the groundwork for developing more sophisticated and reliable AI-based tools for social simulations and psychological research, while also highlighting the ethical considerations of creating realistic AI personas.
https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQD54s8K-T6GkK5v8m-176-13d8o8V5g5Lg4h0gLwLzY5iJ_z38E-qM0h9n8M5vK9f2s-hE4v1gM0D9R9wFh0P0zJ7E7p5r8YwI3J5r-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-.
A Literary Character Approach to Human-like Personality Simulation in …
(2023-10-12) Large language models (LLMs) have demonstrated remarkable abilities in simulating human-like conversations. However, their potential in accurately simulating human personality traits for social science research remains underexplored. This study introduces a novel “literary character approach” to enhance LLMs’ persona simulation capabilities. By constructing detailed “character skeletons” based on established psychological frameworks and enriching them with nuanced, literary-style descriptions, we guide LLMs to generate responses that are more consistent with specific personality profiles. We evaluated our method by comparing the personality traits of LLM-generated texts with those of real individuals from a public dataset. The results indicate that our approach significantly improves the alignment of LLM-generated personas with target personality profiles, as measured by standard psychological metrics. This work lays the groundwork for developing more sophisticated and reliable AI-based tools for social simulations and psychological research, while also highlighting the ethical considerations of creating realistic AI personas.
https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQD54s8K-T6GkK5v8m-176-13d8o8V5g5Lg4h0gLwLzY5iJ_z38E-qM0h9n8M5vK9f2s-hE4v1gM0D9R9wFh0P0zJ7E7p5r8YwI3J5r-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-848-1C-5z-84