Dylan Castillo conducted a rigorous analysis of seven major AI models to determine if labs are deliberately optimizing for 'pelicanmaxxing' (drawing pelicans on bicycles). The study found no evidence that any lab performs significantly better at this specific task than predicted by their general capabilities in drawing animals and vehicles separately. This debunks the notion that AI labs are intentionally training models to excel at this niche meme.
Background
The term 'pelicanmaxxing' refers to the hypothesis that AI labs might be over-optimizing their image generators for the popular internet meme of pelicans riding bicycles. This article discusses a scientific attempt to verify or refute that hypothesis through controlled testing.
- Source
- Simon Willison
- Published
- Jul 23, 2026 at 07:01 AM
- Score
- 5.0 / 10