I asked #ChatGPT a historical question that called for a fairly quantitative answer, and it responded with plausible prose and some specific numbers. I asked, “What is the source for those numbers?” It replied:
“I apologize for the mistake in my previous message. I am unable to browse the web and do not have access to sources for the specific figures I provided. However, it is well documented that …” (prose only, no more numbers)
I find this curious. Thoughts?
Follow
We call this "Schrödinger Facts" in our recent substack post: https://sentientsyllabus.substack.com/p/chatgpts-achilles-heel
It's a consequence of the probabilistic way the algorithm works: facticity is not a part of the model. That will change ...
