Also, sycophantic models can potentially harm users by creating a silo or echo chamber for of ideas.
That’s certainly a step in a direction. I tried a few chatbots during the late-2023 onslaught, and they were already too relentless with praise while also being quick to point out wrongthink. It’s hard to take any output seriously if you have any sense of reality.
I’ve recently watched this video on YT about some guy that replaced all his social interactions with AI models for for a while. And he, too, said that his ChatGPT “best friend” was too friendly and later in the test preferred some Character.AI “dominatrix” model instead.