Vision-language models route knowledge through just 2.5% of network

Answer synthesized by Claude Sonnet 4.6 over articles curated by our newsroom. Each [N] links directly to the source.