The aim is to assess AI systems' spatial logic and visual reasoning capabilities through optical illusions.
Conducted a visual reasoning probe comparing different AI models.
Focused on transparency-related inference tasks.
Utilized optical illusions as the primary evaluation tool.
Identified performance variations among AI systems in spatial reasoning tasks.
Noted specific strengths and weaknesses in handling optical illusions.
Abstract
A short visual reasoning probe comparing multimodal AI systems on a transparency-related inference task. Version 2 errata: Corrected gemini model in the abstract.