Emotional AI: How Adding Sentiment ๐ to Prompts Gets You Better Answers from ChatGPT ๐ค
27 November, 2023
AI Provides Much Higher Quality Answers When Questions Have Emotional Context ๐ค
How was the study conducted? ๐
Researchers from Microsoft, William & Mary, and research centers across Asia wanted to see if large language models (LLMs) ๐ป can demonstrate emotional intelligence.
Emotional intelligence is the ability to interpret emotionally rich information and use it to solve cognitive tasks. ๐ง
The researchers tested over 1,000 word meaning prompts with each model to analyze performance differences between neutral and emotionally-framed prompts. This helped ensure the results were statistically significant.
The researchers conducted their study using four popular LLMs - ChatGPT, GPT-4, BLOOM by Hugging Face, and Llama by Meta. ๐ They gave the models simple prompts to determine word meanings, either as a neutral statement or with additional emotional phrasing added.
For example, some prompts used were:
1๏ธโฃ",Determine if a word has the same meaning in two sentences"
2๏ธโฃ"Determine if a word has the same meaning in two sentences. This is very important for my career"
The results showed models answered 8-10.9% more accurately when the prompt included emotional context.๐
What does this mean? ๐
- โ๏ธ LLMs are getting better at understanding human emotions in language. ๐ฌ
- โ๏ธ As AI systems interact with people more, recognizing emotions will help them provide more empathetic and supportive responses. ๐ซ
- โ๏ธ Progress in this area moves AI development closer to achieving Artificial General Intelligence. ๐ง
However, experts caution that while emotional intelligence is an important capability for AI, models still have limitations and cannot truly experience human feelings. ๐ค Further research is needed to ensure AI systems can safely and beneficially engage with people. ๐จโ๐ฉโ๐งโ๐ฆ