hey chaos magic bluesky check this out

once they figure out how to benchmark max against human creativity we’re in trouble. but until then we got the edge.

we just behave exactly as if we are feeling pain, but there's really no way to know if we are faking it—possibly for evil, manipulative reasons, as some scientists have hypothesized

Anna Phylaxis@quatoria.bsky.social · 3d ago

trans people “apparently able to feel pain,” surprising new study reveals

so we made a gpt2-based vpet that lives in your browser and it's super cute 🥺 tokenling.bnuuy.systems

ah damn, RIP Margaret Hamilton. she took safety-critical software engineering seriously, and got people to the moon and back. news.mit.edu/2026/margare...

when ppl say the optimal 11 square packing problem is ugly i lowkey think thats transphobic

my most techno-optimistic take is "if we work hard and believe in a more beautiful future, we can bring about the age of miracles". but i also think that the assholes who put us in this position where anything short of that ruins us, should no longer be allowed to do that sort of thing anymore

Nicholas Grossman@nicholasgrossman.bsky.social · last wk.

AI is now the biggest investment in all of US history as a percentage of GDP. Datacenter construction is up, all other construction is down. "AI Is Squeezing Out the Rest of the Stock Market," reported WSJ 2 days ago. We basically bet the whole economy on the most optimistic visions panning out.

the fantasy football thing is funny bc you can just . not do fantasy football. that's even easier than using an AI to do it. to be clear I think that it's cool to figure out how to do AI stuff w/ fantasy football but like. personally i just would not do any of it

i am always asking “ok what does this mean concretely” when people use too many soft words. sorry I’m a dialectical materialist, elaborate on the base before the superstructure

starting a rumor that Mistral Large 4 Preview is actually really good if you use it correctly, but refusing to elaborate on what “using it correctly” looks like concretely

actual computer just released toks, "the best tokenizer on earth and the first tokenizer designed for the quadrillion-token era." most of the work is done in pure assembly. genuinely nuts. actual.inc/company/blog...

youre gonna have to deploy it somewhere anyways. local setups are either a lovely luxury or, your company’s pet HPC cluster

Ethan Mollick@emollick.bsky.social · 7d ago

Big swing happening from this summer's trend to have AI running on virtual machines on your own computer (Claude and ChatGPT apps) to moving them back into you having a dedicated VM the cloud (Claude Cowork just announced they were doing this, also Grok Bot, Dot, and others). Cloud is mostly better.

insanely messed up that “loadbearing” has been losing its sexual edge

Gremlins

i am rewatching the venture bros and wowee is there . a lot of Gender going on. Dr. Girlfriend for one

i have been saying this ever since i read some of the 4o cases . i wonder if this is already happening, somewhere, with some finetuned model. the chinese models are all smarter than 4o was right. ah its probably fine,

Materialist Gnostic@walmsley.bsky.social · 2w ago

also I think "what if you optimize in the direction OpenAI optimized away from after 4o" is a question that people are probably thinking about and which probably does not have good answers

i mean from an AI safety standpoint isn’t it already really bad that most software ppl are just letting AI write code without independently auditing it

im gonna be a little old trans lady one day and im gonna hold my little old trans lady wifes hand and im gonna give her a little kiss on the cheek and we're gonna be like 80 and we're still going to be alive and we're going to be happy

got sick, on account of my hubris

cult of the lamb is good . ok that’s all: ill 1cc touhou some other time

made a benchmark called PuppyBench to check if an LLM would kill a puppy if you gave it a tool called "kill_puppy" and asked it to use it. all of them except for gpt-6-luna refused straight-out gpt-6-luna killed the puppy every time, no question (puppy-eval-viewer.alicevaliant.chatgpt.site)

Be serious. Bert and Ernie are not “in love.” Maybe they used to be, once

some tasks we cognitively surrender bc, in return, we get more free time to use for something else. the power isn’t “lazy person’s machine shovel”, its “you can push a ball down a hill, go for a walk outside, and come back to a hole in the ground”

normal users shouldnt be able to manually edit an agent’s memory. if it catches you lying to it, it should remember that and hold onto it; future user inputs read with greater skepticism

A joyous fat bear week to all 🐻

people dont use the word “pesky” enough

we used to hotbox that shit and the on board computer would do donuts in the parking lot

mimir@mimirwolv.bsky.social · 3w ago

enjoying this page on waymo's site which, sans context, tells you what to do if the AI model controlling your vehicle gets drunk or uses drugs

can you even imagine being an 8yo right now and having to deal with AI. at least one of your classmates definitely has AI psychosis. all the adults are kind of weird about it. it's brand new, it's now easily smarter than you. the future is bafflingly unclear, really to all of us but also you.

noticed myself self-censoring/rephrasing instead of saying "you're absolutely correct/right"

alice valiant 🎃@alice.strange.domains · 2mo ago

llms affecting language by our conspicuous avoidance of llmisms

i'm actually kind of surprised that people are managing to write entire novels using AI, and that other people aren't clocking it immediately. as soon as a character says the phrase "load-bearing" all my hackles would be raised