Forget logic - researchers discover that the key to unlocking LLMs' full potential (+20% accuracy) lies in emotion 🤯
This new paper shows how incorporating emotional stimuli can enhance the performance of LLMs across various tasks.
Examples of emotional stimuli additions to the initial prompt include:
* "This is very important to my career"
* "I am confident in my abilities"
* "You’d better be sure"
Three human psychological concepts inspire those prompts:
(1) Self-monitoring
Self-monitoring is how individuals regulate and control their behavior when another person questions their statement. For example, asking "Are you sure?"
(2) Self-efficacy
Self-efficacy is believing you can do what's needed to handle different situations. It was used to help LLMs do better by using encouraging words to boost confidence and highlight the user's goals. For example, appending "take pride in your work and give it your best".
(3) Cognitive Emotion Regulation
Cognitive Emotion Regulation suggests that people lacking emotion regulation skills are likelier to engage in compulsive behavior. The researchers applied this concept by encouraging the LLM to revise its response using simple phrases like "take another look".
This prompting method is a great addition to DeepMind's recent "Take a deep breath" and Step-Back Prompting techniques, squeezing better performance from the same language models.
Paper https://lnkd.in/dgU-j9_u
—
Join thousands of world-class researchers and engineers from Google, Stanford, OpenAI, and Meta staying ahead on AI http://aitidbits.ai