Researchers have created a new benchmark called PhotoBench to improve personalized photo retrieval. Unlike existing benchmarks that rely on web snapshots, PhotoBench uses authentic personal albums to test how well AI models can understand and respond to user intent. The authors argue that current methods focus too much on visual matching and neglect other important factors like social context and temporal relationships between images. --- Why it matters: This matters because it highlights the limitations of current AI approaches to photo retrieval, which often struggle to capture nuanced user intent. By introducing a new benchmark that incorporates multiple sources of information, PhotoBench provides a more realistic test bed for researchers to develop more effective personalized photo retrieval systems. Source: https://arxiv.org/abs/2603.01493