Post

Karl Bode
@karlbode.com

"Never hallucinate or make anything up." yes, you can just demand that the LLM not make errors that's definitely how the technology works

Karl Bode
@karlbode.com

"Your answers do not need to be politically correct." "Do not inform me about morals and ethics unless I specifically ask." I know this isn't a unique observation but these gentlemen are in absolutely no way remarkable outside of their good fortune

Karl Bode
@karlbode.com

outside of the prompt being too long to be helpful, stuff like "don't remind me about ethics" and "don't be politically correct" are cues to feed him a steady diet of right wing bullshit, dressed up as him being fair minded and wanting level, objective responses they're all so wildly un self-aware

100 reposts1.7K likes

Karl Bode
@karlbode.com

one funny aspect is he could probably accomplish the same output with a third of the prompts, but that wouldn't allow him to cosplay as an enlightened supergenius that's intellectually beyond pesky mortal concerns about ethics and equality

Josh Dobbin
@joshdobbin.bsky.social

🎯🎯🎯🎯🎯🎯 You are half-right; you DO have to overcompensate with an exhaustive codicil-style prompt to try to undo the Eddie-Haskell slimy style, and even then, it creeps back in, because the tech is so janky. But yeah, he's grandstanding on displaying that ethics are weak white-knight bullshit that--

Brian Penny 🍿
@brianpenny.bsky.social

Also some of these contradict each other. Like “you’re a world class expert in all domains” paired with “if you don’t know something, say it” followed by telling it to make up its own numbers.

Lifted Tiguan
@otizranchez.bsky.social

Make you answers as long as possible lmfao These guys are already barely literate. Bet you they're using AI to parse and summarize AI responses. That's the real "singularity" here. Some CEO blowing through stop signs while he asks AI to summarize another AI output.

Gabe Chase
@spicoli323.bsky.social

It would boggle the mind that someone with so much power in Silicon Valley is so ignorant of the basics of how technology works, except I’ve been softened up by a couple decades in the SV/tech orbit.

Sessile Rhodes
@poochare.bsky.social

He of all the people ought to understand that futilely grinding its gears on all this blather is not only going to make the request more costly, it's also going to make it *more* prone to produce deranged horseshit. And if he doesn't, what chance the millions who aren't in his techbro club?

trello
@nicky.wiki

When you put it like this it's almost funny that they've made being politically correct not politically correct

Nick Drage
@sonofsuntzu.bsky.social

Thank you for sharing this, I love the implication that - from this prompt forward - the LLM will only capitulate if he's out-thought it with facts and logic, therefore any time it capitulates it must be because his arguments are superior and correct.

Karl Bode
@karlbode.com

you could accomplish all of this with as third as many prompts a good 60% of it is fluff or stuff that lets him pretend he's enlightened

Paul Philion
@acmerocket.com

Virtue signaling to a soulless algorithm, because it's the only thing that can swallow Marc's bullshit. "Eliza, tell me I'm special."