A study by researchers at Microsoft and the University of Toronto has shown that LLMs do indeed have some level of emotional intelligence, and that this intelligence can be enhanced by providing them with emotional prompts.
The researchers conducted a series of experiments on different LLMs, including Flan-T5-Large, Vicuna, Llama 2, BLOOM, ChatGPT, and GPT-4. They found that LLMs performed better on a variety of tasks when they were given emotional prompts, such as “This is very important to my career” or “I’m really excited about this project.”
For example, on a task that involved determining whether two words have the same meaning, LLMs were able to achieve an accuracy of 57% when given no prompt. However, when they were given the prompt “This is very important to my career,” their accuracy increased to 67%.
The researchers also conducted a human study to evaluate the quality of generative tasks performed by LLMs using both vanilla and emotional prompts. The results showed that the emotional prompts significantly boosted the performance of generative tasks, with an average improvement of 10.9% in terms of performance, truthfulness, and responsibility metrics.
The researchers believe that their findings could have a number of implications for the development and use of LLMs. For example, they suggest that emotional prompts could be used to improve the performance of LLMs in tasks such as customer service, education, and healthcare.
The researchers also believe that EmotionPrompt works for LLMs because it helps them to better understand the context of the task at hand. When an LLM is given an emotional prompt, it is able to better understand the user’s intent and the desired outcome of the task. This leads to improved performance on a variety of tasks, including both deterministic and generative tasks.
The sources for this piece include an article in Arxiv.