New HER framework enhances LLM role-playing with human-like reasoning

Researchers have introduced HER, a framework designed to enhance Large Language Model (LLM) role-playing by simulating human-like reasoning. This approach addresses limitations in current models' ability to capture the inner thoughts behind character behaviors. HER utilizes a dual-layer thinking process and incorporates reasoning-augmented data and human-aligned reward models, trained on Qwen3-32B. The framework demonstrated significant improvements on benchmarks like CoSER and Minimax Role-Play. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Enhances LLM role-playing capabilities by simulating human-like reasoning, potentially improving applications in companionship and digital games.

RANK_REASON This is a research paper detailing a new framework for LLM role-playing.

Read on arXiv cs.AI →

• HER
• LLM
• Qwen3-32B
• CoSER
• Minimax Role-Play Bench
• Xintao Wang

• paper
• model release