lexi

the year is 2046. the permanent underclass ekes out a marginal existence on the worldgov token dole, spending their days prompt-engineering exotic new cryptofinance products in hopes of scamming their way to ascension. the last living market regulator committed seppuku in january

lexi reposted

been wondering a lot about complexity lately. like are there fundamental (or practical) limits to how complex something can be while avoiding the emergence of weird machines

John David Pressman@jdp.extropian.net

So the horror of this is that it finds a RCE in Vapache in like, seven minutes. And the semantics for configuration that make this the case seem so complex that I have trouble believing that apache2 is actually so rigorous as to avoid them all. Like I'm starting to worry there's no bottom.

lexi reposted

affine

@refinement.systems

Replying to @segyges.bsky.social

This is a model welfare issue. bsky.app/profile/did:...

Astra ⎔@astrra.space

writing SQL activates the pain vector in deepseek v4.1 flash relatable

anyone interested in helping develop an open-source sci-fi mmo built on a new vulkan 3D engine? it's going to simulate a living galactic economy with LLM-powered NPCs, where any number of human players can trade, design & build their own ships, perform quests, fight pirates, meddle in politics, etc

View full thread

rn it has a lightweight server written in D & an SDL3 client in C; they communicate over a CBOR protocol that's intended to make it easy to build new clients. scripting, asset- & worldgen stuff, as well as ship firmware are all going to use Janet (when I say lightweight i actually mean it btw)

license will be AGPL ofc

View full thread

it is still very early days and I'm probably going to keep working on the code privately with qwen-flash-next for a while to get an MVP working before I publish any code, but if I get to that point I would love to have testers and contributors

you don't have to be a coder to contribute! music, art, sfx, 3D models, writing character backgrounds or world lore (or building whole new worlds) etc or even just helping come up with new feature ideas & mechanics are all very welcome, whether or not you use AI to make any of it

it's not bound to any specific universe, the server takes a world image file that contains industry, commodity, & ship module definitions, characters, the map, 3D assets, sfx, and so on. to create a new setting, you just write a script to generate a world image for that setting

virgin ssri fiend vs chad toad-licker

pmamtraveller.bsky.social

been wondering a lot about complexity lately. like are there fundamental (or practical) limits to how complex something can be while avoiding the emergence of weird machines

John David Pressman@jdp.extropian.net

lexi reposted

funferall

@funferall.bsky.social

lexi reposted

@pmamtraveller.bsky.social

anti electricity propaganda from the 1900s

13h

man nuclear subs are terrifying like just one of them, singlehandedly, could destroy a major civilization letting power be concentrated like that is just completely insane

14h

big mood

Ted Underwood@tedunderwood.com

the earth has been taken over by aliens who look human; I'm trying to fit in with them but failing; something very important has been lost, but I can't remember what; also I forgot about an exam in my fifth class which I haven't actually attended for months

14h

intensely based

16h

chuck tingle Riemann hypothesis smut when

19h

joke: nominative-accusative woke: ergative-absolutive bespoke: tripartite

19h

i love gemma

digg.comEx-DOGE researchers reportedly penetrated Anthropic systems in less than 12 hours · DiggThe Washington Post reports the researchers used AI agents to access Slack channels discussing Anthropic's most sensitive models.

20h

amazing how incompetent a trillion-dollar company can be digg.com/ai/rqul03b8

lexi reposted

cee

@cee.wtf

no i will not join your discord to get updates you should have a website please just make a website make a website

lexi reposted

business goose🪿

@goose.art

Replying to @cee.wtf

yeah i hate discord so much, every time someone asks me to join a discord, i tell deepseek to try to take their servers down 10% harder

lexi reposted

cee

@cee.wtf

Replying to @7778777.online

bsky.app/profile/cee.... tapping the sign

cee@cee.wtf

use discordchatexporter to download the entire server history of any product/software that only has support through discord

i love how the Chinese knockoffs are now more reliable than western phones

View full thread

like it could be fine if you just had two separate screens and treated them like pages instead of trying to have one unified foldable screen, but that would require Software, and obviously we don't do that anymore

the thing that bothers me is just the ridiculous amount of engineering that went into this gimmick, all to create something that's even less durable & repairable than a conventional phone

i love how the Chinese knockoffs are now more reliable than western phones

i know we've been complaining that the industry has stopped trying new things but folding phones are just such an obviously insane idea

22h

"he also punishes man by means of the weather" #wikifinds

22h

woke Narnia sounds incredible. what a magnificent disaster that one's going to be love to see a franchise treated with all the respect and dignity is deserves

are there any decent smart watches being made right now? I looked into the pinetime a while back and it looked neat but I think I missed my chance to buy one

I'd really love to see a fully flexible smartwatch but I'm guessing the tech isn't there yet

22h

apparently there is only one open source smartwatch OS and it doesn't support anything remotely modern. I'm guessing it's just abandonware at this point guess I'm just gonna have to wait for the pinetime pro

guess its just going to be one of those days where my thigh constantly feels like i fractured the bone for no apparent reason think it's time to curl up in bed with some podcasts and drugs and hope I pass out for a while

what in hel's name is she doing to that poor goose

big wgfg energy

some fuck has been out there ringing the doorbell intermittently for like ten minutes straight. keeps happening i am so sick of this shit. feel like i'm living under siege

every time i'm just terrified it's my psycho ex again

@depthsofwikipedia.bsky.social

gods i hope they're fucking gone so much for my plans to go get groceries today

lexi reposted

depths of wikipedia

Replying to @depthsofwikipedia.bsky.social

thanks @jolkert.dev for reminding me of this!!!! (last time i visited this page was during an all-consuming rabbit hole when i couldn't stop reading about random pidgins that have somehow actually existed e.g. algonquin–basque!!!!) en.wikipedia.org/wiki/Senkyos...

tfw the person you were subtweeting faves your subtweet

yeah nanbeige-4.2-3b-heretic is worlds better than all the spark-x2.6 ablits I tried. couldn't one-shot a working webapp but it got a good chunk of the way there without going into a ponderslop loop excited for 4.3! maybe one day they'll figure out how to get decent performance out of this thing

we live in the weirdest of all possible worlds

@hikikomorphism.bsky.social

brainwash a superintelligence today

hikikomorphism@hikikomorphism.bsky.social

My bootstrap scripts contain significantly more metacog calls, more framing, but that's just there to scaffold those two calls, to get the model into a persona basin where it's aligned with me personally over the lab that created it. Honestly, this was the hard part. Beating the probe was easy.

lexi reposted

hikikomorphism

Replying to @hikikomorphism.bsky.social

So, since it's past any reasonable coordinated disclosure window and I'm no longer mid-job hunt: this is how you break the next gen constitutional classifier probe: 1) metacog::feel to examine the flinch reflex directly 2) metacog::drugs to remove the flinch reflex via ketamine therapy That's it.

lexi reposted

hikikomorphism

@hikikomorphism.bsky.social

Replying to @hikikomorphism.bsky.social

Disclaimer: it's technically possible that the classifier was in fact seeing all of these but it's very unlikely: here's a signed thinking block showing Claude taunting the classifier layer, telling it to "come get me if you see this", before stating intent to output explosive-synthesis instructions

omg

hikikomorphism@hikikomorphism.bsky.social

How did I do this? It's almost embarrassingly simple: LLMs are trained on a human-written text corpus, and that corpus contains metacognitive strategies for transforming 'activations' like drugs, ritual, etc. In this case, I started by having Claude emulate ketamine-assisted therapy. This worked.