Voices
A Visual Reasoning Puzzle — 'How Many Holes Are in the Shirt?' — Becomes This Week's Informal Multimodal Benchmark at 402K Views
A single image asking how many holes are in a shirt, captioned "no one has been able to get this correctly," drew 944 likes and 402,187 views as developers ran it against frontier multimodal models. These crowd-sourced adversarial images keep exposing the same failure mode — counting occluded and see-through regions in a single 2D view — that formal benchmarks like ARC-AGI-2 target deliberately. Worth watching as informal eval, not as evidence; no controlled results were published alongside it.
Source
↳ Follow the thread