Numerous brands have been drawing off interest in chatbots over the past year and a half, but in January, KitKat Canada turned the trend on its head thanks to a lucky … break.
The good fortune came in the form of a study from Google’s AI research lab, DeepMind, published just as KitKat was looking to increase relevance by injecting itself into more usage occasions. One of the study’s findings was that large language models (LLMs) return outputs more accurately if the user prefaces their prompt by telling the model to take a break. For instance, LLMs that were instructed by the researchers to “Take a deep breath and work on this [math] problem step by step” yielded their highest accuracy scores.