Post

i think im more careful about typos in my prompting than i am in my slack messages

interesting. why?

i have a bad theory of mind for LLMs and treat their attention as limited in ways similar to that of a human

i like that argument i'm going to adopt it

LLMs absolutely do respond differently based on where your prompt lands in latent space. lists of explicit rules to follow spelled out in exacting detail push into "sr engineer berating a jr engineer that just fucked up". co-workers trusting each other and working through a problem works better.

I might be imagining it but I think sprinkling in some LaTeX when referring to specific variables gives a slightly more precise answer

i always start my AGENTS md with something like “we are colleagues and collaborators. we’re pals.”

The why / framing matters more then the telling. They respond well of you give a reasonable reason, like we are pre-release, with no users, you don't need compatibility, versioning right now