Discovering the Hidden Vocabulary of DALLE-2

Giannis Daras, Alex Dimakis

Keywords: text-to-image, generative models, dalle-2, adversarial examples

TL;DR: We discover that DALLE-2 seems to have a hidden vocabulary that can be used to generate images with absurd prompts.

Abstract: We discover that DALLE-2 seems to have a hidden vocabulary that can be used to generate images with absurd prompts. For example, it seems that ``Apoploe vesrreaitais'' means birds and ``Contarra ccetnxniams luryca tanniounons'' (sometimes) means bugs or pests. We find that these prompts are often consistent in isolation but also sometimes in combinations. We present our black-box method to discover words that seem random but have some correspondence to visual concepts. This creates important security and interpretability challenges.

Student Paper: Yes