Sure, as you certainely knows Llms are not build to tell the truth, they are build to satisticaly generate text.
Everything dépends on model, prompt, température and son many more parameters… like data acces, tools. The purpose of this project is not to write true things, if I wanted to, I could specify in the prompt “don’t invent facts”, I could give it Solid spec, doc, or a link to . It’s like a baby just learning to speak, but does not understand all the things he says… but if you give it the good infos, how to use it, what you exactly want, I think it could be a good tool.
Until now, I had mistakes for long conversation ( when the context is overflow, some infos are dropped). And data was unstructured. Using Solid’s graph as knowledge base was one of my goals since I tried Ai tools, following concepts links, exploring and updating data. The may issue until now was CRUD operations and indexation for speed of retrieving data without following each link, reading each resource everytime user ask a question.
And here is a good starting point 
Then I could visualize the result as a mindmap, plugin it to my 3D-solid-mindmap tool :
