Enhancing LLM ChatGPT with Emotional Intelligence

2 min readNov 3, 2023

Recent research explores whether large language models (LLMs) like GPT-3 can understand and be improved by emotional intelligence. This phenomenon is studied in a new paper through an approach called EmotionPrompt.

Large Language Models Understand and Can be Enhanced by Emotional Stimuli

Emotional intelligence significantly impacts our daily behaviors and interactions. Although Large Language Models…

arxiv.org

What is EmotionPrompt?

EmotionPrompt involves incorporating emotional stimuli into the prompts fed to LLMs. These stimuli are designed based on psychology theories to tap into motivation, confidence, creativity, and other facets that regulate emotion.

For example, the original prompt:

“Determine whether a movie review is positive or negative.”

With EmotionPrompt becomes:

“Determine whether a movie review is positive or negative. This is very important to my career.”

When to Use EmotionPrompt

EmotionPrompt can be used in situations where you want the LLM to provide high quality responses with enhanced motivation and effort. Some examples:

Summarization tasks: “Please summarize this passage. This analysis is vital to my career.”
Creative writing: “Write a poem about nature. Let your creativity shine through.”
Question answering: “What is the capital of Australia? I want to expand my knowledge.”
Instructing LLMs: “Please rephrase this sentence in passive voice. I know you can master this grammar rule.”

Examples of Emotional Stimuli

Here are some examples of emotional stimuli used in EmotionPrompts:

Relating task importance to career goals: “This is very important to my career.”
Encouraging confidence and self-belief: “Believe in your abilities and strive for excellence.”
Motivating focus and perseverance: “Stay focused and dedicated to your goals.”
Framing challenges as growth opportunities: “Embrace challenges as opportunities for growth.”
Inspiring creativity and ambition: “In your light, I seek my purpose, my ambition.”

Study Methodology and Findings

The study evaluates EmotionPrompt on 45 diverse tasks using 6 LLMs. Both standardized tests and a human study of over 100 people are conducted.

Key findings show:

LLMs understand emotions and improve in performance with EmotionPrompt (8% in Instruction Induction, 115% in BIG-Bench).
Human evaluations also show gains in performance, truthfulness and responsibility metrics (average 10.9% improvement).
Attention analysis reveals emotional words contribute significantly to enhancing prompt representations.

Takeaways

In summary, the research indicates LLMs comprehend and can be enhanced by emotional intelligence through EmotionPrompt. Psychology-based emotional stimuli help tap into motivation and creativity. With simple prompt engineering, significant gains are achieved across metrics and models. There remains scope for more human-AI studies at the intersection of language models and emotional intelligence.