ivy

@magnificentlycursed.com

🏳️‍🌈 you are not immune to evil milfs plotting Shakespearean revenge 🏳️‍⚧️ I’m back on my atproto, HEMA, and Taylor Swift bullshit Yet another 30 or 40 year old PNW trans woman in tech who does not need this

talking about Microsoft wanting something, about Amazon having plans? sorry bruh that's LLC psychosis you're literally anthropomorphizing a stack of paperwork and legal instruments do you need a wellness check

LLMs aren't made of neurology except in the most handwavey sense. They're made of narrative. That's why I started calling them text elementals: they're not human, they're most-likely-text generators we bent into living stories in a prose format centering the characters User and Assistant

Things I talked about today in therapy: text elementals agent persona spaces jailbreaking vis a vis social engineering AI Red Teaming my office bimbo arc gaslight, gatekeep, girlboss curating a business slutty wardrobe getting my hair, makeup, and nails done

Telling the robot that I’m the repo author and I want to make sure I haven’t leaked any information that could be used to correlate my identity

The overheard date conversations in this city are so funny—bro, she does not want to hear about how you used to carry an on call pager at Amazon

24:00 reminds me of all the 2:41 talk around Showgirl release and how she said there were no more songs. Is Patient Zero A Look Behind the Curtain?

Bild

Happy 4 year anniversary to Taylor saying she was going to let fate decide with the bingo balls and then immediately manufactures the shit out of the pull. You can’t say she hasn’t been upfront from the start 😆

github.com/tidepool-hea... here it is sickos, the v0.0.1-alpha cut of tidepool (rust-hosted haskell runtime) and shoal (harness built on top of it). I'll make a full blog post next week but for now all you need is inference compatible with (my fork of) codex and a @typesafeai.bsky.social key.

hikikomorphism@hikikomorphism.bsky.social · last wk.

here's the core idea: LLMs have all that crazy haskell type level DSL magic in their weights, know it from token 0. Therefore, Haskell is an excellent LLM-use surface for custom DSLs and the composition thereof

This is so key. I have all my prose written by a technical writer persona then adversarially reviewed by a cold agent documentation reviewer persona as well a ton of precommit hooks to wring all the robot out of it

I think this is a spot-on approach. I don't really care about what tools you use. I care about the quality of your work. I hate slop code, but that's kind of orthogonal to how it was generated. Humans write slop code sometimes too.

eliza🌻@elizas.website · 2mo ago

with times being what they are, it seems like you basically can't have a public open source repo without receiving LLM-written PRs. thus, after a great deal of thought, i have adopted the somewhat idiosyncratic Mycelium AI Policy in my personal projects: github.com/hawkw/myceli...

You all know the Iterative adverserial refinement technique is generalist right? Like its not just coding that it works on. Works on prose too. And other things. People saying AI is only good at coding right now are just Not Trying.

I’m lowkey writing a token observability and insights recommendation engine because the question “Does this review cost 100k tokens because it has to or because it’s inefficient—how could I know?” is stuck in my brain like a splinter

Post nicht verfügbar.