Post

This haunts me constantly as I use LLMs, and I find it so fatiguing. You're constantly balancing between 'this thing is inhumanly good at making useful connections with data' and 'this thing has no idea what the connections mean or if they're real'. That's a strange thing to navigate and verify

the ability these things have to both make idiotic mistakes and catch things I never could have noticed is truly something

Which means you actually have to pay attention and consider the outputs. You can't 100% reject or accept whatever comes out. Since these things are unblocking magicians, you don't get much of a break. It's relentless review and verification Pardon my whining