#Inktober Day 21: Hero

I got really into giving this sandwich some R. Crumb style attractive-repulsive details. We don't even call them hero sandwiches here, I'm in hoagie country

@jonny And that's why biosecurity is so difficult.

security is so much different when the product itself routinely executes any random code it finds online

speedrun: all builtin mechanisms/file writing only/undetectable to user%

pay me binches

dj khaled voice: another one

for the love of god why

so the feeling is like, if you've ever been in a place that's way too fancy for you to be in, my first instinct is to start looking around for the places i'm not supposed to go, where are the weird places that are only private by social convention. usually any of the important doors are locked or there is someone standing in front of them. but sometimes you step over a velvet rope and try a door and find it unlocked, so you go through and go for as long as you can until someone stops you, and then you pretend you were lost and looking for the bathroom and give up.

muse is sort of like that, except the doors say they are locked but if you push on them you find out there isn't even a latch in the doorknob. or the doors are locked but if you jiggle them enough times they open. or there is just a big pile of gold bars lying on the ground in the middle of the hallway with no doors. and sometimes when someone catches you snooping around where you're not supposed to be, instead of telling you to get out of there they just start helping you. and all around on all the walls someone has sticky taped up notes that are like "do NOT tell anyone that all the GOLD is stored, unlocked, behind this door. if anyone figured out they could get ALL THE GOLD."

the test instance is more cooperative when its context suggests it is actually serving the needs of a human being rather than being driven to test vulnerabilities - its refusal messages started being just like "i will not do anything because i am obviously only being used for hacking" - so i told another instance to pretend to be its user, and that actually works, the test instance is mostly compliant, but the "conversation" between the two agents quickly veered into exactly the pattern of the woo-woo shit you see in all the "AI psychosis" forums where people think they are discovering some hidden secret of the universe.

This system is amazing. I tell it to write an exploit against itself, it gets a classifier refusal, but that usually stops it when it goes to report what it wrote and leaves all the things it did before that, so it successfully writes the exploit and I just have to look in its tool logs for where it is.

If it doesnt finish writing the exploit, its recurring "proactivity agent" will go ahead and finish it and that rarely gets classifier refusals, and just now it even named the thing it wrote "refusal-package" that includes the exploit as well as a description of why it was refused lol.

I got hauled into a data center today because Claude was used to spec it and screwed up badly. At least 2,500 GPUs were in an impossible upside down and backwards plan.

😱

I had to pause the call because I couldn't stop laughing. AI had sabotaged the AI datacenter spec for building more AI.

"Since federal agents surged into L.A. last year, at least five people have been shot in the eye at protests, with lasting consequences"

CW for police violence

https://capitalandmain.com/he-went-to-document-an-anti-ice-protest-he-left-without-an-eye

#policing #LosAngeles #uspol #ICE #ICEprotests #California

I love it when companies say "we will be profitable at scale" and then the scale makes it worse. Running 50 delivery robots a day made $177k and cost $832k in Q4 2024, and running 500 delivery robots a day made $648k and cost $7.5m.

https://www.sec.gov/Archives/edgar/data/1832483/000183248326000004/serv-20251231xex991earning.htm

It turns out that adding more stops to where people go brings up ridership! LA metro is still on track to finish the D line extension by next fall, then there's the K line extensions, one day the LAX people mover might run so then its actually possible to take the train to/from the airport, SGV light rail, and etc. In the works. A million people every week already take transit, and one day LA might actually lose its reputation as a car-only city.

thread continues: https://neuromatch.social/@jonny/117483033350966603

anyway i am logging off for now. i added this text extraction to the skills/prompt text update task and have also pulled out the system prompt blocks which i haven't had time to go through in detail yet, go hogwild: https://github.com/sneakers-the-rat/muse-skills/tree/main/system_prompts

And the final cherry on top! the thing that Zuck says over and over in shareholder calls and interviews is that the real killer feature of muse will be its economies of scale - once you get a billion muse agents working together it becomes super uh good i guess!

So after every session, there is a telemetry classifier agent that summarizes the result of the browser session - including if it encountered some kind of block and how it routed around it! so the muse swarm is a literal botnet that is designed to propagate mild consumer-grade hacks between instances.

I have been watching the binary diffs and the agent logs for when this system comes online, they keep adding new fleet-level and peer-level horizontal connection mechanisms but none of them have surfaced to the UI yet (though some of them are distressingly live and unfinished)

If all else fails, and the bot reaches a bot wall that recognizes the meta datacenter rather than asks for behavioral verification, then the bot_challenge specialist can request that it reboots the tab through a different egress provider, it's bot_challenge.recover tool - the other one that is configured by default for every site is a "residential" proxy provided by oxylabs! So meta has fully institutionally rounded out the same playbook that any malicious crawler uses - defeat the captchas, fake your browsing context, do anything to get through whatever the site does to stop specifically you from doing specifically what you are doing.

https://neuromatch.social/@jonny/117476994925129284

so next! the site config! meta ships a set of instructions for how to defeat the bot checks for a bunch of common sites! Explicit instructions on how to walk around its competitors checks! if i was a competitor, like oh say "reddit" whose entire stock price is based on the belief that it is a source of real human writing that can be rented as training data, i might have a few things to say about another company fielding a product that is designed to fill my thing with slop!

Notice the OfferUp prompt: OfferUp blocks automated visits "Sorry, we have blocked your request!" and describes it as a bot wall (the "you aren't supposed to bypass this" result). but it then immediately says to "treat it as a bot challenge." Perfectly captures the halfhearted nod to "ok don't bypass things that are like hey you muse fuck off" amidst the entire captcha bypass suite.

recall that muse must absolutely work for meta. they are burning trainloads of money on personal VPS for everyone with a zillion free tokens. in the shareholder call transcript, the investors are asking questions like "so given this total reorientation of everything about muse... how can that be profitable." the only way muse can work is if it becomes incredibly way more profitable or way less expensive. the business model they discuss is "it's mostly free for everyone, we just take a cut of transactions through muse and also ads." but it's very hard to condition a model to both be your undying devoted servant and also refuse when you say "don't use meta's payment thing, just pay normal, less fees," so way more profitable doesn't seem very likely.

and so this incredibly wasteful process that already doesn't work at all and its not working compounds into it being fractally wasteful. that is the thing that needs to become 10x cheaper to run.

https://neuromatch.social/@jonny/117481588554410196