I'm stuck using Copilot for work, and for 3-4 months I began every prompt with 'Do not comment or offer an opinion on my query' and eventually Copilot stopped with the sycophancy.

Now I'm actively working on deprogramming myself of the notion of speaking to it as a human and making a point to treat it like an old-school MS-DOS ish OS and give it commands instead of being conversational, because it's important to remember that this fucking thing is a computer program.

Absolutely. I must use Claude at work. I put instructions in my CLAUDE.MD to, "lay off the sycophantic bullshit. No praise or extraneous details, just deliver answers."

It doesn't work, per se... But it seems to reduce the attempted fellatio, at least a little.

I noticed Copilot doing this and I told it to stop that behaviour through the customization prompt. "I don't want platitudes, flattery, or an AI that agrees with me. I need an answer with sources and nothing else." seems to have fixed its worst tendencies for now.

Now that I am forced to use it for work I at least get output that's usable. The only thing that still irks me is when it attempts to speak from experience ("In my experience...") and when it attempts to humanize itself ("Personally..."). I haven't managed to kill those behaviours yet, but they are getting really annoying.

I don't use AI, mostly because I just don't see a use case for it personally, but the ass kissing is definitely something I don't enjoy either.

Being surrounded by sychophants can be a very bad thing.

I think anyone who isn't an American hates that, and probably even some of them. For Gemini at work I've got a custom gem set up with explicit instructions to cut that shit out and just be as factual as possible.