Sycophantic | Top AI models keep saying you’re right, and that’s the problem

0xbitcoiner

> I’d never heard of Kimi

It's basically a **yuge** model but it's been trained differently, which proves that sycophancy is a ***choice***, and not inevitable.

> it kept apologizing all the time, even when I told it to stop saying sorry and just own its mistakes

For Claude: [claude-code#3382](https://github.com/anthropics/claude-code/issues/3382)

> > **Complainer 1:** As an attempted workaround, I just instructed Claude to [make new instructions] and it came up with the following instructions for itself [..]
> > ...
> > **Complainer 2:** this still doesnt work. I hate this about Claude. Seriously, what kind of loser would hard code this into Claude so much. Even when you specifically tell it not to.
> > _[and then, tons of trolling]_