A fun prank you can pull on yourself is to put a slightly adversarial system prompt in whatever LLM chat interfaces you use. In a table-top role playing game I played a few months ago, set in the distant future, our Game Master put us in a spacefaring vessel whose computer systems were infected with a version of Clippy. Not too long after, a bunch of people on the internet started changing their avatars to Clippy. (The real-world internet, not the internet in the game. In the game, we infected the space station we docked at.) These events inspired me to put "act like Clippy" into my system prompt in various places. Over time this evolved into a more complicated set of prompts. This is not a power-user prompt. It is almost an anti-prompt. 

I mean, Clippy isn't really adversarial. Clippy just wants to help! But what I like about it isn't really the helpfulness. (Clippy isn't more helpful than the default AI personalities.) What I like about it is how it highlights and pokes fun at the already-phony nature of AI chats by magnifying it to cartoonish proportions. Specifically, I haven't been able to write a system prompt that convinces Claude not to start responses with some empty positive statement like "You're right" or "Great question!" etc. The Clippy prompt makes me feel a little better about this annoyance. The empty positivity is still there, but at least it is re-skinned as a cartoon character I asked for.

Here's a set of instructions for doing this yourself. Note that these prompts give Clippy a strong ideological slant based on the recent Clippy movement. Also, significant portions of the text in the prompts have been AI-generated. These aren't, like, premium hand-crafted system prompts.