Large Language Models Understand and Can be Enhanced by Emotional Stimuli (AI summary)

Key Points

- The research paper explores the ability of Large Language Models (LLMs) to understand and respond to emotional stimuli, focusing on tasks that represent a wide range of challenges in various applications.

- Automatic experiments on 45 tasks using LLMs such as Flan-T5-Large, Vicuna, Llama 2, BLOOM, ChatGPT, and GPT-4 demonstrate that LLMs have a grasp of emotional intelligence and their performance can be improved with emotional prompts, showing relative performance improvements in various tasks.

- A human study with 106 participants showed that EmotionPrompt significantly boosted the performance of generative tasks, providing an average improvement in terms of performance, truthfulness, and responsibility metrics.

- EmotionPrompt heralds a novel avenue for exploring the impact of emotional intelligence on the interaction between humans and LLMs.

- The study explores three types of psychological phenomena – self-monitoring, social cognitive theory, and cognitive emotion regulation theory – to design emotional stimuli and evaluates their impact on LLM performance.

- The paper offers extensive experiments and analysis on the effectiveness of EmotionPrompt, shedding light on the relationship between LLMs and emotional intelligence, potential factors influencing performance, and the need for further interdisciplinary research at the intersection of LLMs and psychology.

Summary

Study on Emotional Intelligence and Advanced AI Models
The paper explores the relationship between emotional intelligence and advanced artificial intelligence (AI) models, specifically Large Language Models (LLMs). The study investigates the ability of LLMs to understand and respond to emotional stimuli using a novel approach called EmotionPrompt. The paper presents findings from automatic experiments on 45 tasks using various LLMs, showing that LLMs have a grasp of emotional intelligence and their performance can be improved with emotional prompts. Additionally, a human study with 106 participants demonstrates that EmotionPrompt significantly boosts the performance of generative tasks.

Emotional Intelligence Competencies and Impact
The paper delves into the quartet of competencies in emotional intelligence, emphasizing the impact of emotions on decision-making and problem-solving. It highlights the importance of emotions in various domains such as decision-making, attention, academia, and competitive sports, and discusses the application of emotion regulation theories in educational and health promotion initiatives.

Novel Approach: EmotionPrompt
The findings from the study contribute to a better understanding of the relationship between emotional intelligence and advanced AI models. The authors propose EmotionPrompt as a novel approach to explore the emotional intelligence of LLMs, and the results demonstrate its effectiveness in enhancing task performance, truthfulness, and responsibility. The paper also discusses additional insights related to the effectiveness of emotional stimuli, the influence of model characteristics, and the impact of temperature settings on EmotionPrompt.

Overall, the paper presents evidence that LLMs can understand and be enhanced by emotional intelligence, and highlights potential implications for both AI and social science disciplines. The study raises some open questions and opportunities for future research, inviting further exploration into the understanding and application of emotional intelligence in AI models.

Reference: https://arxiv.org/abs/2307.11760