The craziest part is probably the "2x decrease in false statements". The whole thread is an interesting read, though, that illustrates LLMs are not as good as you think they are.
aifaq.wtf
"How do you know about all this AI stuff?"
I just read tweets, buddy.