Sensitivity of Generative VLMs to Semantically and Lexically Altered Prompts

Open in new window