The ChatGPT service that's all the rage is OK at some basic wikidata queries: https://twitter.com/erik_paulson/status/1599491203866558464 (first 4 chats) https://twitter.com/erik_paulson/status/1599491207247187968 (next 3)
It's a terrible search engine so it doesn't know any q numbers, but I was surprised at how it knew some properties to use - it must have found enough example queries on the web to know things like start date and end date. I may have been cheating just a bit when I told it to use P642, and I'm not sure why it came up with trying to use a country as a constraint, but it's directionally correct.
Those first two tweets were my second try at it. My first try I might have used a bit too much SPARQL in my feedback chats and it felt a bit like writing the query for it but I was equally impressed - it even used a blank node in its query https://twitter.com/erik_paulson/status/1599481401178730497 https://twitter.com/erik_paulson/status/1599481404307697665
I don't know if it remembers my previous chats, so I don't know if my second time with a chat session was preconditioned on the first, but even if it did the first time I tried it was really good.
Admittedly a simple query but it was still impressive. I think remembering that it's a prose generator and you want to steer it towards writing you the right thing (and so pointing out its errors) gets you quite a long ways.
-Erik