Here's a great example of how it handles uncommon medical questions.

Simple question, can trans women get endometriosis? This question is great for this because common sense will say, "no, of course not," and LLMs are nothing but distilled common sense. But common sense is often wrong. Common sense assumes far greater difference of the sexes tham actually exists. Here's what copilot has to say:

Copilot goes for the wisdom of the crowds, and the crowds fail. It even tries to respond thinking I'm talking about trans men rather than trans women.

And yet, here is a case report, clear as day, of a trans woman getting endometriosis.

https://academic.oup.com/jes/article/7/Supplement_1/bvad114.2078/7291745

And this phenomenon has also been observed in cis men undergoing estrogen therapy for prostate cancer.

I admit this isn't a common thing, as far as we know, among trans women. But the AI flat out says the possible is impossible.

I think AI aids the American right specifically in their quest to eliminate the role of experts (with their "liberal" educations and "progressive" values such as basic human decency for all, etc.) in discussing and determining policy, especially in regulating industry.

They want to replace protesting scientists, archivists, historians, philosophers, and other "whiny" experts with a machine that appears to make logical arguments and to be an expert, and is much more malleable and agreeable.

You see this on X all the time. Some expert says a thing and some Nazi looks to grok to refute it.

Does it matter that none of this changes reality? Not really to them. Because they aren't looking to be correct, they're looking to exercise power without pushback.

Sounds interesting. Would you recommend the book to someone who is working with Deep Learning and thus understanding the technology pretty good? I am a bit tired of reading from "AI experts" that do not know the technical capabilities and limitations of technology they are addressing.