In 2022, Ethan Mollick, an AI researcher and University of Pennsylvania professor, found himself needing to amuse his daughter on a boring plane ride. For some help, he turned to what he knows best and began prompting Midjourney's latest model to generate an image of an otter—his daughter's favorite animal. Taking inspiration from their current predicament, he then asked the model to imagine the otter stuck on a plane. The resulting images "were all horribly distorted," he recalled.

The memory stayed in his mind, though, and he has continued to throw the prompt at new generations of image models, creating a casual but fast and effective way to measure their advancements. Lately, AI models have gotten quite good, and for the newest models, he's upped the challenge to really test them.

"You have to basically say, 'Do an image of otter scientists explaining the otter-on-a-plane benchmark with images of otters on planes,'" said Mollick, who has become a widely followed AI guru and wrote a 2024 bestseller on understanding the technology, "Co-Intelligence: Living and Working With AI."