While I agree that AI talking like that is totally unacceptable, isn't that example of Grok's stupidity at the end kinda useless as an example? Mixing text with images and logic is one thing that (all?) current image generation models are very bad at. I asked ChatGPT to generate the same thing ("create an image of five different basic geometric shapes and label each one"), and it made essentially the same errors as Grok.
https://chatgpt.com/share/67b2a77b-adf4-8009-8b91-99321c6dad...