Hey, so I responded about it in other thread, but let me respond here.
Most challenges were created by us, but we use LLM for example to generate comments in the code or generate some simple boiler plate.
Sometimes, it is also useful to create similar challenge to already existing one.
https://shaderacademy.com/shaders/glsl/ranked_1/fragmentExpe...
But I still stand by the ai-gen'd aspect