3 ms·
How did you get an LLM to be snarky or did you do something else?
by PaulHoule 11mo ago
How did you get an LLM to be snarky or did you do something else?
- pizzathyme 11mo agoSame question and great work. I would love to know the prompt details of how the hacker news truth was captured
- nottorp 11mo agoYes, this is absolutely brilliant! Teach us the prompt, o great wizards!
- PaulHoule 11mo agoTo answer my question myself I gave Microsoft copilot this prompt: I want you to rewrite this headline "Amazon will allow ePub and PDF downloads for DRM-free eBooks" into something a little humorous and snarky that reveals the underlying truth that would bring a wry smile to tech-engaged but big tech-skeptical hacker news readers. This has to fit in the 80 character limit for Hacker News so keep it appropriately short. Also I want you to reply with exactly one headline and not anything else so I can use your output as part of a processing pipeline and i get the response Amazon Finally Remembers eBooks Aren’t Supposed to Be Prisoners which I think is great. I started with the first paragraph and got something too long with some explanation. I added the second, and got three replies and more explanation. The three replies were all "good enough" in my mind but added the third paragraph to control the output.
- nottorp 11mo agoI prompted Gemini to tell me how to prompt itself to get similar results on other news sites and it said I should give it a description of the intended audience and what it finds funny/snarky. Which looks like what you did.
- doomspork 11mo agoYou can prompt it to do so. Look you "Persona based prompting" as a great and fun example of controlling what the LLM spits out and its tone.
- smokel 11mo agoLet's try to get a story on little Bobby Tables on the front page and find out.
- minimaxir 11mo agoModern LLMs are now actually good at having a sense of humor.