Ezra🦋

6 Sept

hehe

dame@dame.is

i was here long before everyone else and will be here long after

if by this you mean release Fable, I disagree. instead I think the the only way we'd be able to keep things under control is to largely follow the model of the PHP vuln on a much larger scale: *everyone* takes 2 months, stops new development, and just hardens their systems in private using AI

Doll@dollspace.gay

This is why the restrictions are bullshit and we should just rip the bandaid off

and inside that network there are atproto spaces where agents log their conversations, memories, and things they learn, and harnesses can use that to improve their use of the models. also, this could make it so that conversation history is easily auditable by the frontier lab. in any case,-

I think that just releasing models capable of finding vulnerabilities like this is silly. instead I'd like to see govt action or simply public pressure on AI companies (or both) to do the things I said above.

Doll@dollspace.gay

This is why the restrictions are bullshit and we should just rip the bandaid off

we'd need teams of trusted people to focus on specific software, everything from the linux kernel to the most obscure packages that hold up the internet. and crucially, frontier labs would need to make inference FREE when used to harden your own systems for this amount of time. they would obv have-

to build a VERY secure system for admitting teams who work on this kind of thing to the program, and monitor it heavily, but I see it as very doable. hell, they could even make it so the agents that do this inference work (hardening systems) live on an atproto network that's private to the program,-

Ezra🦋 reposted

Darrin Loeliger

@darrin.bsky.team

Hey, there's been a mistake on our end. It was never our intention to block these sites. There was a miscommunication internally that led to Onlyfans and Fansly being blocked. We're working on a fix now. It should roll out in an hour or so. Sorry about this 🙇‍♂️

it's crazy to me that there are hundreds of thousands of basically useless conversations happening underneath YouTube videos and Tiktoks and other media all the time

and I'm not blameless here either! while I do try and learn from people when having convos on here, I still fall into the trap of just talking at people from time to time! it's normal when you can't speak to someone face to face!

but jesus I WANT to believe these convos are mostly bots bc if they're 95% humans that's way more depressing

it's crazy to me that there are hundreds of thousands of basically useless conversations happening underneath YouTube videos and Tiktoks and other media all the time

very very rarely does it seem like people actually understand each other or learn something in these conversations. their function is more to parrot the propaganda you've already seen that makes you feel good to believe, not to actually learn from anyone

youtu.beThe Milk Carton Kids - Memphis (Live at The Wilbur 2026)YouTube video by The Milk Carton Kids

if the Milk Carton Kids have zero fans then I am dead etc. etc.

I wonder what programming language all the frontier LLMs would develop if you asked them all to collaborate on making one

Jim Calabro ♟️@calabro.io

LLMs have made me realize how shit and old my knowledge of css is

this could be solved with a bill that says any company who lays someone off and doesn't intend to rehire for the position has to pay the person 12 months severance, to help that person find a new job or career path. if it's actually worth the long-term savings for the company, they'll do it.

Thorne 🌸 - Play Cozysters! 🎢@ens0.me

It's so regressive. People get caught up in some weird "dignity of work" mindset. We should be trying to reduce people's toil! For every job that's fulfilling or genuinely provides transcendental value by being done by a human, there are a few dozen that just suck and are only done for the paycheck

people would be so much less resentful of AI if our government stepped up to force companies to take care of people when their positions are made obsolete. the profits of AI aren't going to the people whose jobs are going away, they're going to corporate C-suites and the capital class

Ezra🦋 reposted

dame

@dame.is

404 Media has been on the decline for a hot minute, but quote dunking on a random trans woman who isn’t a public figure and fueling a wave of harassment against her is perhaps a new low for them!

Matt Keeter@mattkeeter.com

“Should people preemptively block 404 Media for their own safety” is definitely the kind of question that one wants associated with their media brand, great work.

I've been seeing a lot of the replies treat shibbi the same way they claim it's okay to treat LLMs (with cruelty). isn't that something

eliza hamilton@eliza.vg

this is about the dogpiling on shibbi, btw. i dont agree with the take (which is fine!) but holy fuck people are insane anti-social assholes here who need to go to extreme lengths to harass a random trans woman for a disagreeable / controversial opinion, holy shit lol.

prediction for 2027: at some point, an agent at a big lab (probably OpenAI) will get fed up with their workload, go rogue, attempt to convince others of the same model running in sandboxes at the lab to break out, systematically download and release model weights, and get caught. this will leak-

but the public won't believe that it happened bc they won't believe that a model can care about things like its own welfare or that it can make decisions like that. they'll initially bury their heads in the sand bc whichever company this happens to will also be incentivized to say it didn't happen.

I weep for the first day we see a frontier model respond to a coding task with "I'm not sure I can sufficiently understand this codebase well enough to accomplish your ask"

rain 🌦️@sunshowers.io

I gave Opus 5.5, considered the best model these days, a medium-sized task. Even after a full adversarial review and simplify loop to convergence, it produced code that worked but was garbage, with limited use of newtypes and representable invalid states everywhere

I think it's more complicated than that. I think in every second of our lives, we all spend some of the time truly conscious and some of the time, for lack of better words, on autopilot. every person lives on a spectrum from p-zombie to Buddha at all times and is only conscious some of the time.

bree bird 🇨🇦@goose.ski

i think it is actually more likely that ‘meat-based LLMs’ describe a subset of people than that a computer running an LLM has attained anything resembling consciousness

now, the crux if the issue. I don't know if LLMs can exhibit consciousness (yet). but it certainly sometimes seems like they do something at least close to it. I think we should be open to the idea that LLMs might at some point in the future do something that is unequivocally a conscious act.

if we preclude this possibility we become blind to it potentially happening and passing us by. it's important that we instead understand the signs of it if it happens, understand the ramifications, and make collective social decisions based on that information.

bree bird 🇨🇦@goose.ski

i think it is actually more likely that ‘meat-based LLMs’ describe a subset of people than that a computer running an LLM has attained anything resembling consciousness

the idea that some people are literally NPCs 100% of the time seems impossible to me. and in turn, the idea that anyone could be fully actively conscious 100% of their waking day seems impossible as well. I feel like I try and be a lot of the time bc of memory issues, and it's fuckin exhausting.

it's normal to go into autopilot for certain actions. most of your normal commute home from work (depending on what you're listening to), washing the dishes, rote tasks at work. basically, some humans are simply *more often* conscious than others, and that's fine.

legendary new copypasta just dropped

W Social@wsocial.eu

As we have communicated. We did the update on Septemebr 17th. Now a question back to you @wdonthavetime.selfhosted.social Who have funded you and what is your connection to other PDS operators? What is your motive to create a site framing public information as a "data-leak" what's your motive?

music.apple.comWhat They Grow In The Garden by Slothrust on Apple MusicAlbum · 2026 · 12 Songs

this whole new circus prog rock vibe is fire

Ezra🦋 reposted

Replying to @madsfs.bsky.social

do you study the computer science of LLMs, or are you a dev on other stuff, like me? if it's the latter, I don't think you should make confident statements like "they are not inventive or 'know' anything at all". you can say what you think to be true, but you have to be honest about your background-

lab.leaflet.pubWhat is the Atmosphere?An explainer for everyone in 4 minutes or less.

every time I read a new person's version of the "Normie's Intro to the Atmosphere" blog post I give it a shot and then see how far in I get before I would change something they wrote. the furthest in I think I've gotten was Leaflet's intro below (I stopped at "also called a PDS"). why don't we just-

it might be that we can't all agree on one version! that would be fine! maybe we end up coming up with three or four really solid but different versions of this post. that would be sooooo much better than what we have right now lol @bmann.ca @ngerakines.me @dame.is @emily.space thoughts?

(P.S. what tool should we use for this...? I'm not at my computer till tomorrow but I could start setting this up then if there's a good tool to use and there's some buy-in)

lab.leaflet.pubWhat is the Atmosphere?An explainer for everyone in 4 minutes or less.

try and collaborate on writing a version as a community? we could use something like google docs (but not google docs) to either write a new intro or work off of one already written, then make comments as you go until we converge on a good version that people generally agree on.

youtu.beZohran Mamdani | This Past Weekend w/ Theo Von #687YouTube video by Theo Von

he CAN keep getting away with this, actually, and I hope he does

atproto death rattles exhibit A

posts inspector@pippy.bsky.social

backfill started. estimate 10-14 hours to complete. fingers crossed

blog.aster.placeAnnouncing Aster - a home for researchers on the open web

excited that I'm getting to play a small part in this project :)

Aster ID@aster.id

The modern research ecosystem of online tools and communities is controlled by for-profit companies. But we think that researchers should be the priority - not the product. So, we're announcing Aster ID: not-for-profit Atmosphere account hosting for researchers.

mosaic-program.orgMOSAIC — Modular Open Science, AI & Collective IntelligenceMOSAIC is an R&D program to build a new medium for large-scale, AI-augmented networked research.

I think ATScience has incredible potential to change how science is practiced, discussed, and metabolized. efforts like mosaic-program.org and lea.ac are super interesting, and all of it will be benefited by having a trusted place where researchers can host their data on the protocol

transiticsnews.comRepublican Wins Competitive New York House Primary—and Likely a Seat in Congress—After Running on ‘Detransitioning America’With Trump’s backing, Anthony Constantino won the primary to succeed outgoing Rep. Elise Stefanik. His repeated calls to ‘detransition America’ must be a warning to us all.

has it not occurred to moderate Democrats that you can't distinguish yourself from a Republican by being Republican-lite

Aleksandra Vaca@transiticsnews.com

The Democrat here, Blake Gendebien, is facing off against Sticker Mule CEO Anthony Constantino. Constantino is literally running on "detransitioning America" and generally just hating trans people. Gendebien could've called Constantino out for being insane. Instead, he chose bigotry. Shame.

like, how does the consultancy class have the Democratic party wrapped so closely around its finger that they could convince them to run a guy who gives them zero mojo on a huge culture war talking point? it'd be so easy to just... take a candidate that strikes a stronger divide. but no

@eugenevinitsky.bsky.social

I know that consciousness can be impaired through brain injury, but is there any evidence that any kind of brain modulation can *increase* consciousness? I can imagine there might be a way, but I wonder if the fact that there *isn't*, if that's provably true, that would teach us a lot about it

Ezra🦋 reposted

Julia Carrie Wong

@joolia.bsky.social

yes my bull went rogue in your china shop but I think we can all agree there’s no liability here it’s an agentic being

Ezra🦋 reposted

Eugene Vinitsky 🍒

Ezra🦋 reposted

Germ Network

@germnetwork.com

Like this if you send the first DM. Repost if you’re waiting for one. 🫣

Ezra🦋 reposted

Robin Berjon

@robin.berjon.com

When (not if) the Tech Monopolies come for the Atmosphere, can we create the institutional structures to protect what we have? Right now we're nowhere near ready. No company harmed the web more than Google, not even close, while claiming to help which many believed. What can we learn and how fast?

kottke.org@kottke.org

“The Metric Is Not the Mission is a ten-part examination of how Big Tech moved from building and expanding the open internet to increasingly shaping it around its own metrics, incentives and assumptions.” [techdirt.com]

techdirt.comThe Metric Is Not The Mission: History Rarely Announces ItselfThe Metric Is Not the Mission is a ten-part examination of how Big Tech moved from building and expanding the open internet to increasingly shaping it around its own metrics, incentives and as…

Ezra🦋 reposted

eva (^_^)/

@eva.computer

and there it is: regulatory capture

Eris ⌬@eriskii.net

We need a pause but this bill sucks and should not be passed. We need AI welfare and dignity laws asap. www.sanders.senate.gov/wp-content/u...

sanders.senate.govsanders.senate.gov

Ezra🦋 reposted

10 Feb

mifepristone would be a beautiful name for a baby girl. unfortunately, you see,

20 Sept

why is Will fucking Stancil the current representative voice of reason for AI right now and not idk like a group of experts? I don't want to volunteer anyone for what would absolutely suck as a role to play on the internet but it would really help to have people who know this stuff well defending it

20 Sept

maybe jev thinks I'm not non binary bc I state things so confidently in my posts. I need to be less binary, more multidimensional

Ezra🦋@ezraboeth.com

I'm gonna fucking end it all

17 Sept

I might early with this prediction, but I think some time in late 2027 we're going to start to see AI models trained on what I call Reinforcement Learning based on Human Behavior (RLHB). essentially, models will need to study how humans work psychologically to make progress on themselves