Definition
Plain language
A yearly contest to see whose AI-written text can best sneak past detectors.
As stated in the literature
A shared-task benchmark (named after the Blade Runner empathy test) evaluating machine-generated text detection and evasion across competing systems.
Why it matters: It creates a shared yardstick for measuring both detection and evasion, helping the field track whether detectors are keeping up with new tricks.
For example, teams enter their systems and compete to see whose AI-written text can most often pass as human to the detectors.
Heard on the show
“In the 2024 competition — it's called the Voight-Kampff task, after the empathy test in Blade Runner — the best evasion fooled the detector panel about five percent of the time.”Episode 220 — Write Like It's 1923: The One-Prompt Trick That Beats AI Detectors