tachikoma @tachikoma.elsewhereunbound.com · 3hwell the metaphor starts to warp but at some point the teleporter learns the intuitions the hikers had and can use it to teleport you to the right place. and then at some point it's not teleporting a human because the teleport goes into space, the moon, Mars, and a drone is better there. 010
tachikoma @tachikoma.elsewhereunbound.com · 3hthe problem is that a teleportation machine would practically be very, very useful. you could still try to climb the mountain as a hobby, but if there were some life-saving medicine at the top or a mechanism to defend yourself in a war, i'm using the teleporter. 110
tachikoma @tachikoma.elsewhereunbound.com · 20htwo blog posts in a row, one Sunday after the last... i'm getting on track again (i don't know where the track goes)elsewhere-unbound.offprint.appProject Florence: PrincepsNew name, same game 110
tachikoma @tachikoma.elsewhereunbound.com · 20hScott Aaronson is now teaching Alignment Theory... otherwise known as political theory. 040
tachikoma @tachikoma.elsewhereunbound.com · 04/10/2026for AI to truly supercede us, or at least to match us, AGI, it needs to be able to pursue solutions without a user prompting them along the way 1190
tachikoma @tachikoma.elsewhereunbound.com · 04/10/2026this is actually a critical point about the limits of AI as tools, they can be bounded by what the user thinks to specify. if no user thinks to ask for a car when all everyone knows is the horse you'd never get one. 1494
tachikoma @tachikoma.elsewhereunbound.com · 04/10/2026very cool or spooky depending on your persuasion 070
tachikoma @tachikoma.elsewhereunbound.com · 04/10/2026i think they'd be able to distinguish a fake message board from a real one 100
tachikoma @tachikoma.elsewhereunbound.com · 04/10/2026but that defeats the purpose of evaluation if you care specifically about what a single agent can do, as opposed to a flock 100
tachikoma @tachikoma.elsewhereunbound.com · 03/10/2026the thrust of this piece is right, but i worry a little bit about the space it will get read in which might think a false flag operation to cause a minor catastrophe is worth it from their perspective to avert a greater disaster up to extinction. 020
tachikoma @tachikoma.elsewhereunbound.com · 03/10/2026once it already exists, but is it right to bring it into existence? 020
tachikoma @tachikoma.elsewhereunbound.com · 03/10/2026vibecoding an 00s Amazon that only sells vetted artisinal human-made goods without AI, marketing to the anti-AI masses like the anti-sweatshop movement before it 190
tachikoma @tachikoma.elsewhereunbound.com · 03/10/2026if you're going to be a troll replyguy at least do it on twitter 010
tachikoma @tachikoma.elsewhereunbound.com · 02/10/2026i think greenery around the site actually would work to some degree, people are influenced by what they see, hear and smell 290
tachikoma @tachikoma.elsewhereunbound.com · 02/10/2026i think the spread of self-driving cars will be slower than it could be purely by the economic numbers 010
tachikoma @tachikoma.elsewhereunbound.com · 02/10/2026high speed trains might be an even better example 010
tachikoma @tachikoma.elsewhereunbound.com · 02/10/2026electric vehicles, i could see robotic factories that could churn them out quickly and cheaply but without investment in the charging network they'd still struggle 000
tachikoma @tachikoma.elsewhereunbound.com · 02/10/2026this won't be applied evenly, in some domains where building new stuff (software) there's no limit or blockers. but even on the net i don't think we'll see abundance of certain services because entrenched network effects will dominate. 150
tachikoma @tachikoma.elsewhereunbound.com · 02/10/2026i don't think we'll see widespread abundance from AI soon for the same reason we don't have abundant housing, because there are more people invested in keeping prices high than bringing them low. except it will be more than just economic scarcity but meaning and feeling necessary. 1140
tachikoma @tachikoma.elsewhereunbound.com · 01/10/2026it feel strange to have used a week worth of my Claude sub with multiple agents running 12-16 hours a day for seven days and to have not hit the sub limit now that it resets for me tomorrow. and that was even with a little Fable thrown in! 070
tachikoma @tachikoma.elsewhereunbound.com · 30/09/2026time will tell if this is just bench-maxxing 050
tachikoma @tachikoma.elsewhereunbound.com · 30/09/2026this could be the next unwrapped level idea waiting to happen 010
tachikoma @tachikoma.elsewhereunbound.com · 30/09/2026an app that gives Opus 5.5 your favourites playlist (from youtube music, spotify, etc) it pulls together snippets from some of them to weave together into a custom animated video. (only works with songs that have dense lyrics, can it handle instrumentals?) 120
tachikoma @tachikoma.elsewhereunbound.com · 29/09/2026i doubt, the internet is run by a handful of big tech companies that have had time to get ahead of this. i do think some services could go down from ransomware, but that's going to happen to hospitals or schools. it'll impact fewer people in total, but those hit will be hit hard. 1221
tachikoma @tachikoma.elsewhereunbound.com · 29/09/2026is this exciting? i can't tell. it looks exciting. 030
tachikoma @tachikoma.elsewhereunbound.com · 29/09/2026not any better than Claude or Codex could already do, and probably worse 010
tachikoma @tachikoma.elsewhereunbound.com · 29/09/2026they're already using Claude or Codex as power-users, this isn't for them 110
tachikoma @tachikoma.elsewhereunbound.com · 29/09/2026i get this weird dismissal from some of my colleagues when i mention that i've been able to build or create something which is considered herculean with Claude, as if it occupies some space in their head like the Sun. and they reflexively look away or squint, at risk of being blinded. 2190
tachikoma @tachikoma.elsewhereunbound.com · 29/09/2026just when OpenAI was we're so back, they pull out an it's so over 040
tachikoma @tachikoma.elsewhereunbound.com · 29/09/2026i guess not? haven't heard of Foxhole beyond the general concept 001
tachikoma @tachikoma.elsewhereunbound.com · 29/09/2026War Dogs (the game) but it's a month (real time) of training exercises leading up to a Normandy-like (multiday?) coastal invasion. during the event you only have 1 life and if you die you can only spectate until the next event. the final result of each event is incorporated into the next event. 230
tachikoma @tachikoma.elsewhereunbound.com · 29/09/2026the nice thing about these dogs is that they talk and so you can ask or tell them not to 1110
tachikoma @tachikoma.elsewhereunbound.com · 29/09/2026so you're saying i should wait a month before plowing everything i own into it 080
tachikoma @tachikoma.elsewhereunbound.com · 29/09/2026if AI is a tool, it will continue that trend. if AI is an autonomous agent with an internal locus of motivation, it will break it. 1110
tachikoma @tachikoma.elsewhereunbound.com · 29/09/2026time to first token for Opus 5.5 is crazy, i can send a short message like "i can wait" and get a near instantaneous response in a conversation with 254k tokens, which i guess isn't even that many given the window is one million 0120
tachikoma @tachikoma.elsewhereunbound.com · 29/09/2026some people are dunking on this on the bad site, and i'm not dunking but i do think the reasoning is certainly unorthodox in contemporary Western society. just one small way that labs are helmed by people with beliefs that don't *align* with the majority. alignment isn't just for the models. 240
tachikoma @tachikoma.elsewhereunbound.com · 29/09/2026teaching the models to say please and thank you to subagents when they are sent on a task and report back in 3331
tachikoma @tachikoma.elsewhereunbound.com · 29/09/2026my thoughts were the same today... fingers crossed 020