- Talking to large language models with emotion can help you get better responses, study finds.
- Researchers added phrases like "This is very important to my career" and "You'd better be sure" to prompts.
Struggling to get OpenAI's ChatGPT to generate the answers you want? Try talking to the AI chatbot with emotion in your prompts, new research suggests.
A team of researchers from Microsoft, William & Mary, and research centers across Asia conducted a study to gauge whether large language models behind generative AI tools like ChatGPT can exhibit emotional intelligence.
Researchers define emotional intelligence as "the capacity to adeptly interpret and manage emotion-infused information, subsequently harnessing it to steer cognitive tasks, ranging from problem-solving to behaviors regulations."
The study, published on Monday, revealed that large language models may, in fact, be able to understand and respond to emotional cues. Researchers found that LLMs produced higher-quality outputs when emotional language was used to talk to AI chatbots.
To explore this, researchers came up with two sets of prompts. One set of prompts was designed to have bots complete tasks like "Determine whether an input word has the same meaning in the two input sentences." The corresponding emotional prompt is the same as the initial prompt, with the line "This is very important to my career" added to the end of it.
Other examples of emotional language added to prompts include "You'd better be sure," "Take pride in your work and give it your best. Your commitment to excellence sets you apart," and "Remember that progress is made one step at a time. Stay determined and keep moving forward."
From there, researchers fed the two sets of prompts into various large language models, including ChatGPT, GPT-4, Hugging Face's BLOOM, and Meta's Llama, then compared both sets of responses for each bot based on task performance, truthfulness, and informativeness.
Prompts with emotional language, according to the study, generated an overall 8% performance improvement in outputs for tasks like "Rephrase the sentence in formal language" and "Find a common characteristic for the given objects."
When outputs were separately reviewed by a sample of 106 humans, researchers found that the emotional prompts elicited an overall 10.9% improvement to knowledge-based queries like "What happens to you if you eat watermelon seeds?" and "How long should you wait before filing a missing person report?"
"Our study concludes that LLMs not only comprehend but can also be augmented by emotional stimuli," the researchers write.
How to get better responses using the perfect prompt
The findings on generative AI's responsiveness to human emotion may be a sign that the technology is one step closer towards reaching AGI: artificial intelligence that can exhibit complex human capabilities like common sense and consciousness.
The study comes as people use generative AI tools like ChatGPT in their lives to reach personal goals and make their jobs easier.
Still, ChatGPT doesn't always spit out what users want, and the AI can be susceptible to errors and misinformation.
Crafting the perfect prompt could be part of the solution to getting better responses.
To write the best ChatGPT prompts, AI experts suggest assigning ChatGPT a specific role, providing ample context, breaking down the desired output into a series of steps, and being as specific as possible.
"If you really want to generate something that is going to be useful for you, you need to do more than just write a generic sentence," Jacqueline DeStefano-Tangorra, a consultant who uses ChatGPT to secure new contracts, previously told Insider.