aster

as someone who was in the AI/LLM mines for work against er will, i have thoughts about the misconceptions around prompt engineering. prompt engineering, contrary to popular opinion, is not simple or easy. also it is far stupider than popular opinion begins to realize.

people think it's equivalent to giving an order for what to do and then the computer does it, well or poorly, depending on its own capabilities to execute the order.

what it actually is, is like attempting to herd a very dumb cat¹ by cajoling it into giving you something close to what you're looking for.

providing it things you think it'll maybe-probably-hopefully pay attention to or take into account, which it sometimes does and sometimes doesn't by some totally inconsistent and arcane metric that is utterly beyond you.

trying one thing which gets a mediocre result every time, and then one incrementally different thing which gets a perfect result but only once.

wondering why it comes when you call like 8% of the time and how to consistently get that more up to 15%--which you can't, because the inner machinations of its little "brain" are on a completely different wavelength to you.

and all those hours you've spent "engineering" by replacing prompt wordings of "should" with "must" with "please for the love of fuck", you could've just asked an actual person who understands words to do what you wanted in about 10 minutes.

1. except cats have real brains. you get me

Happy to announce that my short comic is included in Thank You For Joining The Algorithm, an anti-AI anthology celebrating human creativity in the face of the dreaded machine overlords. It's full of fiction, poetry, a few comics (like mine!), art, and even a logic puzzle! Pre-orders on the website are open now -- it's less than $20!