EiCAP: Beyond Fluency, Probing and Improving Emotional Intelligence in LLMs via Psychologically Grounded Multi-Turn Dialogue
Researchers have developed EiCAP, a new framework to assess and enhance the emotional intelligence (EI) of large language models (LLMs). The framework includes EiCAP-Bench, a 3,174-probe evaluation suite, and EiCAP-SFT, a large corpus for supervised fine-tuning. Experiments show that standard conversational fine-tuning does not improve EI, but targeted EI-grounded LoRA training on Qwen-2.5-7B-Base significantly boosts performance across all EI subcategories. AI
IMPACT This research provides a method to imbue LLMs with emotional intelligence, crucial for sensitive applications like mental health support.