Can Vision-Language Models Count? A Synthetic Benchmark and Analysis of Attention-Based Interventions

ArXi:2511.17722v3 Announce Type: replace Recent research suggests that Vision Language Models (VLMs) often rely on inherent biases learned during