A new study looks into whether large language models (LLMs), including Flan-T5-Large, Vicuna, Llama 2, BLOOM, ChatGPT, and GPT-4, can really understand and respond to emotional cues like humans do.
The researchers conducted experiments with different tasks to see how well these models perform. They found that LLMs do have some grasp of emotional intelligence, and their performance can be improved when given prompts that include emotional stimuli. For example, they saw an 8.00% improvement in one task and a 115% improvement in another.
The researchers also did a study with people to evaluate how well the models perform in tasks that involve generating content. Their findings show that prompts with emotional elements significantly boost the models’ performance. The researchers suggest that this approach, called “EmotionPrompt,” could open up new possibilities for better interaction between humans and AI models.
