A student told me the funniest thing about image-based #LLMs. Ask them for a glass of wine full to the rim. They can't do it (image generated with Gemini). Despite their potential, LLMs can't go beyond what the data they were trained on. Let the game begin! What other images can't they generate?