>The AI can nonetheless deliver bullshit better than a child, despite the child having potential to eventually do better. Honestly, I always felt slinging bull

>The AI can nonetheless deliver bullshit better than a child, despite the child having potential to eventually do better. Honestly, I always felt slinging bullshit was a super useful skill. I'm pretty good at it and synthesizing on the spot from practice. When the big models first hit and I saw how much better at bullshit they were than I was, it really made me feel uncanny valley uncomfortable. Their output had the same truthy+confident timbre that I could write, but it drew from an unfathomably wider shallow pool than I could no matter how much I've read and could tenuously link together. I wish this wasn't the case. I also wish bullshit wasn't a useful skill and more people were inoculated against it. I'm just waiting for the new machine gun phishing wave to spring up where enough PII is available from hacks, and for-sale datasets to craft very persuasive, targeted phishing attacks at scale.