Could of be that Claude 3.5 say this image already during training? Those models are trained on a large body of text and images from the internet, e.g. reddit,