"The last couple GPT-4o updates have made it too sycophant-y, working on fixes"

I noticed last week it had suddenly turned from the Terminator into Jimmy Fallon.

It went from "as a language model I do not have opinions or feelings" to "wow, what a GREAT question, I LOVE that you asked me this, it's so interesting!!".

We all know it doesn't have feelings. It's so phony and annoying, and isn't it a waste of compute? I assumed it was on purpose because they determined people prefer an AI that kisses their ass.

That is a surprisingly honest admission. It is one of the under/un-adressed problems with the "algo" based social media platforms, at some level you realise it is hacking your limbic system but you can't really do anything about this except disconnect. Like a magician using object permanence to fool your brain into seeing things that aren't happening.

Huh. My impression was always that the vendors saw these things' nauseating sycophancy as a feature.

Yes I did the update and now it gushes approval at everything I say.

but the market for sycophancy is exploding!

Darn, I thought I was just asking one incredibly insightful question after another.