† k͎i͎k͎i͎ †

im just a foid with a phantom wallet

affirm

@m00nlitmak 🙁❤️‍🔥

honestly if someone i hate told me the sky was blue i would probably argue that it was actually red and they didn’t know what they were talking about and then call them a fascist even if it made me look dumb in front of the entire groupchat.

@LagxSol lol ok

@kikixsol im hot

i wish hot people liked hiding their identity as much as ugly simps do

@bbarbiedollbody unblock me on tg hoe

ok so here’s what i’ve actually been building, in plain english. everyone’s worried about whether AI lies. but nobody had a clean answer to a weirder question underneath it: when you pressure an AI into saying something false — bully it, trick it, even retrain it — did you change what it says, or what it knows? turns out those are two different things, and you can measure the gap. we ran the same experiment across four different ways of corrupting a model. push on it socially, plant fake context, catch it silently caving, or literally rewrite its weights. and the same pattern shows up: the attack captures the model’s answer, but the knowledge underneath usually survives — and you can get it back by asking again from an angle the attack doesn’t control. the lie lives in the frame. the belief lives deeper. the wildest part is the weights experiment. if you brute-force retrain a model to believe a lie, you can do it — but the model pays for it, broadly. it gets measurably dumber at everything, stuff the retraining never touched, to below random-guessing levels. you can’t surgically overwrite what a model knows without wrecking it. we’ve now shown that on two different companies’ models, two sizes, same result. why this matters: the entire future runs on whether we can trust what these systems tell us. this is a step toward instruments — actual measurement tools, not vibes — that can tell whether an AI’s answer reflects what it knows. the tools are public. anyone can pip install styxx and re-run our numbers. every claim ships with a machine-checkable certificate. and the part i’m most proud of: we found a flaw in our own flagship result last week, retracted it ourselves, republished the corrected version within a day, permanently, under my name. no one caught us. we caught us. where it’s headed: papers are deposited with DOIs, arxiv is next, and we’re inviting outside labs to try to break it. one person, one 8GB graphics card, receipts for everything. if it survives contact with people who aren’t me — that’s when it gets real.

@mouse6900 i wish everyone would stop telling me what to do im easily confused

@kikixsol put the milady on

@mouse6900 this was the next thing in rotation

@kikixsol so?

@mouse6900 but they got mad at me for wearing a milady

@kikixsol the momo

@mouse6900 which one

@kikixsol your pfp is dog shit

@drizzytardio i dont even have nudes it’s crazy what they’re doing with AI now

@kikixsol Based

they really need to stop letting retards have so much money

@idontknowdana @angeIiki i have an extremely funny story about the blood brothers that i cannot say on the timeline

@kikixsol @angeIiki circletakessquare but I did see alexisonfire with the blood brothers one year and that rocked

Reminding fine shit that I used to be a scene queen when I was a 16 so he KNOWS I am not washed

@seasaint stop incelmaxxing

CT women

@idontknowdana @angeIiki alexisonfire or circletakesthesquare

@angeIiki free before the march of flames scene or isetmyfriendsonfire scene?

@0xMatt66 milady

@kikixsol Gmilady

@mikadontlouz tendies already launch fam tf u mean

Good coins launching right now. I like it

@idontknowdana I honestly have not checked up on her in about 15 yrs 😭😭😭 I’m pretty sure she went to dental school

@kikixsol wtf kiki kanibal up to now? john hawk???? crazzzzy memory unlocked!

kiki kannibal survived 2005 myspace and stickam drama you’re gonna get through this queen and if you’re lucky someone will make a youtube series about you youtu.be/VORIXLfq9Ws?is…