Swords Around a Throne is a sign that you're definitively on the path. If your French is reasonably on point, track down a copy of Infographie de l'Empire napoléonien.
Bartrowhatever (almost certainly mangled) may be tough; I have a feeling he may have missed the census taker dying in 1890, and moved in from who knows where.
I did find a Bartrosz name arriving from Pomerania around the right time, so that's another possibility
Upstream of this is teaching professors/teachers how to responsibly use AI. A big part of my academic librarian day is convincing professors to see AI tools as a multitool, rather than just a "thing that writes/cheats." The rise of things like Undermind help somewhat
My first prompt was too vague and it just compared 2000 and 2024. Second try, again with Agent Mode, went smoother. I'm not a subject expert, but the sources seem reasonable given what I know about data generally.
As someone at a university who talks about what products to use/try out on the free tier, this makes ChatGPT a challenge. A lot of inexperienced users are going to get a bumpy ride; I get reliable results on Plus since I can turn on thinking and more thinking, but...
Prompting ChatGPT to research something and put information into an image seems like a frontier it can't really cross. Anyway, all these pictures are more accurate than this one of Chester Alan Arthur that Grok just made me. But then, how many Chester Alan Arthur vectors can possibly be out there
And second, it shows that the models don't sense that this sort of thing is a challenging problem and needs to switch from a nano system to something else; for a product like GPT-5 (which is essentially, to the consumer, a bundle of models) this is a task it needs to do well
