

The images aren’t generated by the LLM part of Grok, they’re generated by a diffusion image model which the LLM is enabled to prompt.
And of course they can create things that don’t exist in the training set. That’s how you get videos of animals playing instruments and doing Fortnite dances and building houses, or slugs with the face of a cat, or fake doorbell camera videos of people getting sucked into tornadoes. These are all brand new brainrot that definitely did not exist in the training set.
You clearly do not understand how diffusion models work.



Politically underdeveloped take. The thing about trying to come up with a third way (not capitalized) is that it will inevitably either result in putting yet another layer of makeup on capitalism and ultimately changing absolutely nothing, or rediscovering some variety of socialism that some guy invented a hundred years ago. Socialism almost by definition has a monopoly on any and all forms of fair and equitable society.
There’s a lot more to socialist politics than communism or leninism.