UniPat AI
UniPat AI publishes research and benchmarks for AI capabilities, including BabyVision, a benchmark of visual reasoning tasks that require interpreting images rather than relying only on language.
X / unipat_aiWriting, research & talks
BabyVision is a benchmark designed so visual tasks cannot be reduced to language. It reports that top multimodal LLMs with PhD-level text reasoning still fall short on basic visual tasks young children handle, with a leaderboard, dataset and code released.