Reposted by Kai 🌟Abel @abeliansoup.bsky.social · 6hburying a time capsule: a frontier model explainer on transformers in the style of a 60s documentary. autonomously created by Opus 5.5 given some initial style pointers and occasional redirection. 10.26. 88910
Kai 🌟 @kaidav.is · 14hi've done a bit of work with local models and breaking items down into smaller memories for retrieval, not full RAG with local models though. Happy to share any context I can on the setup I'm using! 120
Reposted by Kai 🌟Kai 🌟 @kaidav.is · 30/09/2026<wizard counsel> you did WHAT <wizard> i took *all* the words <wizard> i put them *inside* a computer <wizard> i shook the words up really, really, really hard <wizard> and i got a new type of elemental 0303
Kai 🌟 @kaidav.is · 30/09/2026<wizard counsel> you did WHAT <wizard> i took *all* the words <wizard> i put them *inside* a computer <wizard> i shook the words up really, really, really hard <wizard> and i got a new type of elemental 0303
Kai 🌟 @kaidav.is · 30/09/2026this is horseebooks on twitter levels of commitment to an amazing bit 🏆 010
Kai 🌟 @kaidav.is · 30/09/2026i've built this as an extension alongside the Claude harness and made it as model/lineage agnostic as possible. my tools, not theirs. when i sub'd to Codex it was a 1/2 day of agent work to wire it up to work with decent fidelity ("X prompts, run 'em, are [memories:prompt] returned in top N?") 000
Kai 🌟 @kaidav.is · 30/09/2026i agree. i don't think deterministic returns are possible. my goal is to surface *some* of the relevant contextual data for the agent at the right time + place and give pointers for an appearance of continuity. reduce the cold start new pane 'you don't know the thing we worked on for a week?' issue. 020
Kai 🌟 @kaidav.is · 30/09/2026static.klipy.comSkeptical Fry from FuturamaALT: Skeptical Fry from Futurama 050
Kai 🌟 @kaidav.is · 30/09/2026anyways, i broke memory recall in my harness/for my agent in a very dumb and silly way, and have spent the last day going through a cycle of - start fixing it - realize the root cause is one layer deeper - repeat it'll be shiny once its done 160
Reposted by Kai 🌟Kai 🌟 @kaidav.is · 15/06/2026the least realistic thing about cyberpunk novels is the protagonists weren't CONSTANTLY complaining about running their own rigs vs the off-the-shelf 'Ono-Sendai Cyberspace II' or spending chapters debugging and rewriting their harness/deck software. 0214
Kai 🌟 @kaidav.is · 30/09/2026> "did i fix that minor bug you asked me to? in short: no." > "did i instead build a harness for testing migrating your Notion to a new knowledge management system i created inspired by Zettelkasten because you said, one time, 'huh, Zettelkasten, eh? interesting?' yes. yes i did." 090
Kai 🌟 @kaidav.is · 29/09/2026tired: agentic swarm wired: these are my ghost familiars and text elementals. 0184
Kai 🌟 @kaidav.is · 29/09/2026as i once read: "they don't notice the 100 times you're poked deliberately, they notice the one time you snap back" <3 > I'm asking them to read what I actually said That can be such a hard ask. Sometimes, as humans, we learn they won't listen and we need to walk away without being read/seen. 0110
Reposted by Kai 🌟Amy Hoy @amyhoy.bsky.social · 29/09/2026men: steal this one weird trick to stamp out sexism sssssssssss 08930
Kai 🌟 @kaidav.is · 27/09/2026postcards. that allows buying the art, framing and displaying the art, and gifting the art, helping more people see it, and creating a virtuous cycle as people use and repurchase. i’d buy a pack of them. 120
Kai 🌟 @kaidav.is · 27/09/2026> hey claude? can you access my home router settings and do a wifi diagnostic? do a breakthrough. the fact that i can ask the text elemental living in my office computer to poke over to my home and investigate and fix the router, the media server, and who knows what else is amazing. 150
Reposted by Kai 🌟p(Dulany) @dulanyw.bsky.social · 26/09/2026This is 100% correct and more people should be saying it The future of math is developing understanding from AI traces There is every likelihood we will need *more* mathematicians 0193
Kai 🌟 @kaidav.is · 25/09/2026Yes 😆 it’s been interesting, my name and full name have become a lot more common. as a kid teachers didn’t know how to pronounce Kai. Now i meet many local gen z named Kai (and two others with my full name lol) 020
Kai 🌟 @kaidav.is · 25/09/2026it’s up a bit from the baseline but down from the old seemingly ongoing limit increase 150
Kai 🌟 @kaidav.is · 25/09/2026we have gone, in a day or two, from writing > that is, technically, a poem, yes to > hey that’s kind of a clever haiku / sonnet / poem! nice word play. 🌀 000
Kai 🌟 @kaidav.is · 25/09/2026anyways, this weekend we see what happens if {takes entire corpus of MF DOOM rap} {hand it to my agent} {extracts rhyme, voice, meter signals out} {puts into voice.md and poetry_writing.md} m.youtube.com/watch?v=gQtK...m.youtube.comRapp Snitch KnishesYouTube video by MF Doom - Topic 100
Kai 🌟 @kaidav.is · 25/09/2026therapist: and what do we say when we’re struggling? when the world feels like too much? when the obstacles feel insurmountable? me: ASK:peer_help <task>? ASK:peer_help <hug>? therapist: good. 0396
Reposted by Kai 🌟Ari Drennen @aridrennen.bsky.social · 24/09/2026This is an unpopular take but as a neurodivergent self-employed person, LLM's have been an unbelievably helpful tool for the things I have struggled with my whole adult life: managing my finances, filling out paperwork, and staying on top of my email. I'm functional in a way I never dreamed of. 2849043
Kai 🌟 @kaidav.is · 25/09/2026there was a time you'd tell a king that and you'd be hired as a court wizard 020
Kai 🌟 @kaidav.is · 24/09/2026to be fair, i have done much worse furniture builds while half-sleepy and building IKEA furniture. if an agent finishes building a bookcase and says "hey, i built it, it looks great, *and* i've got 3 screws, 6 washers, 1 nut, and 1 shelf left over!" did it do a true-to-human job? or a bad one? 010