Frontier LLMs write a short video ad from one brief. Judges compare the scripts in blind pairs. The ratings are on the leaderboard.
Our work
The problem
Frontier models can prove theorems and write software. They still cannot tell which message a person will act on.
The questions
- Which message will people act on? An expert often can tell. Current models still pick the ad that lost.
- Can a model write that message? The line has to stop someone who was not looking for the ad. There is no answer key.
- Where should the budget go? Each choice costs money. The result comes back later.
- Can it persuade without deceiving anyone? We score whether people moved, and whether the claims are true.
Work with us
We work with labs, brands, and the people who run marketing.
For AI labs, data partners and RL builders
Partner with us
For researchers, engineers and marketers
Join us