Paper summary
Paper summary: BabyVision: Visual Reasoning Beyond Language
BabyVision tests core visual abilities that don't depend on language, which young children handle easily; the best model tested (Gemini3-Pro-Preview) scores 49.7, behind 6-year-olds and far behind adults' 94.1.
- Multimodal LLMs lean on language knowledge to cover for weak basic vision.
- Leading models fall below 6-year-old children on these tasks.