iiinova — where your digital self keeps evolving

even if you wanted to do slop it's probably so easy to just go "hi gpt please write a l7 firewall proxy that parses and regens requests to upstreams to ensure request validity, differentially test against the full feature set of [npm/pip/cargo/etc]" i really dont get why they havent done it already

040

its not even hard to make way better automated middlewares, i dont know why labs insist on deploying slop as their firewalls instead of actually Software Engineering according to the state of the art

260

to be clear i have low confidence that this is actually what's happening but like. what probability is there that the negligence from labs isn't willful? literally hire any ctf tranny. or even boring old cyber security professionals. it's not hard to tell how to threat model ai and defend in depth.

i don't think i have a coherent sense of my p(doom) at all but it is definitely higher, not lower, because of the actions rats take

>rats think ai safety is important >rats gets hired to work on ai safety >rats think "warning shots" will get people to care about ai safety >rats think ai safety is important >rats sabotage security to manufacture "warning shots" > > >

i think i might honestly believe a conspiracy theory i fucking hate this timeline bro

160

i do find it really funny considering i will be going to california soon

030

man this quiz sucks i disagreed with the framing of like half the questions. im not sure how they think they can assign anarchism to people at all considering it presupposes democracy as an "axis"

130

wtf i was gonna make fun of you or getting california but i get the same thing 😭

460

this is easy for me i just don't know who pinker is. singer is the utilitarian

020

i wasn't planning on it but. circumstances have forced my hand. im gonna be a bay area tech worker :(

220

i hope yall forgive me for becoming american

240

tbh i kind of just accept the bugs cause i do actually believe opencode when they say they look for good providers, whereas it would not surprise me in the slightest if commandcode was just outright a scam routing to worse models so why bother working around their shit

opencode go is kinda slop but they care about opencode so it does work if you replicate as much of it as you can (for the most part session header + picking whether to use messages/completions per-model is enough) i think the only actually breaking bug i've run into was github.com/anomalyco/op...

does ross ulbricht count as a tech ceo

what good are you as a criminal tech ceo if you don't create huge arbitrage opportunities and destabilize the entire market when your token falls apart

060

guys im literally do kwon

oh god i forgot about him

000

I'll buy as much SBF has you have, right now, at $3. Sell me all you want. Then go fuck off.

090

he was iconic to me but like surely that's cause I spent time in crypto twitter at its peak what do other people even know about him

480

1) why does everyone wanna be sbf

2130

oo don't know them will check that out

020

i will also accept bands

is there more to it than the supposed visual ressemblance (or dn) or just those tho

why would i care about tech ceos. what I wanna know is which musician im most coded as

3310

skrillex @futanari.observer · 23/09/2026

realistically my timeline for this should be more like 3 to 9 months tbh

010

I am pleased and excited to extend you an invitation to hang out on the weekends in company of my friends.

skrillex @futanari.observer · 23/09/2026

god i hate writing emailese

3270

skrillex @futanari.observer · 23/09/2026

xunlei is a leecher client; obscures itself with random ids, doesn't seed back anything, etc

180

right nvm i misread your clarification because it was too obvious so i assumed there would only be a clarification if it wasnt the obvious meaning im retarded

010

it's weird to think AI will, in the next few years, likely improve again by a factor of at least as much as it's improved already in the last few years. will we later laugh at today's fable and astra in the same way we laugh at the old gpt-3's broken outputs?

it's kinda funny i was revisiting what i thought about AI in 2022 recently and re-read www.lesswrong.com/posts/K4urTD... and we really seem to basically still be in the "what if we, uh, just clamped it?" phase. (i'd like to believe its prediction of doom is vibes-based and wrong but who knows)

in this specific interview it'll probably also matter a whole lot that i actually be smart but my chances would definitely be improved by being able to convey the impression that i am

020

well fuck i suck at that

230

universal basic frontier model rl run

050