Clio Health First Deadline

Gimme AI Break: KitKat Finds Tech More Reliable After a Rest

Even cutting-edge tech needs a breather to recharge its batteries

Generative AI is just like us! Its answers are more accurate following a little time off. Google DeepMind unearthed this potent factoid, which KitKat put to the test across work that just broke in Canada.

For a brand with “Have a Break” as a tagline, this stuff practically wrote itself—after a much-deserved breather.

Video Reference
Have AI Break, Have A KitKat

“A creative from found a published study from Google DeepMind that included deep data around giving AI a ‘breather’ or a break before asking it to process a question or a task,” says Joel Holtby, co-founder and CCO of Courage, the agency behind the campaign. “We then spent time trialing and testing within AI platforms while recording the data that came back, which aligned with the Google study.” 

A traditional summarization request of Twelfth Night yielded 56 percent accuracy while a request to take a break then summarize the play resulted in 78 percent accuracy.

The agency tested this theory on basic questions to those more in-depth.

“We spent months trialing different questions and tasks against various language based AI models,” Holtby tells Muse. “As far as craziest questions asked, we went pretty far and deep from recipes to deep theories on the existence and proof of space and time.”

Clio Health First Deadline