Sign in

Deep Learning State Machine

@collapsedsquid.bsky.social
175 followers 312 following 253 posts
PostsRepliesMedia
Deep Learning State Machine @collapsedsquid.bsky.social · 09/10/2026
RUnfair that Yudkowsky gets to be the prophet of AI when Arjen Lucassen was clearly the one with real foresight
020
Deep Learning State Machine @collapsedsquid.bsky.social · 06/10/2026
I say we need to put Games Workshop in charge of keeping brits healthy and safe from crime
120
Deep Learning State Machine @collapsedsquid.bsky.social · 03/10/2026
Weird alt history takes on China, Russia or the Ottomans are what Kaiserreich is sort of about, Germany winning WW1 is the headline point, it sort of has to be because that's where the major world powers are.
011
Reposted by Deep Learning State Machine
ae @aelkus.bsky.social · 01/10/2026
what did he know
730352
Deep Learning State Machine @collapsedsquid.bsky.social · 30/09/2026
Hear stories from the dot-com days about how the IT department was frequently the only profitable department from charging other departments of the company
020
Deep Learning State Machine @collapsedsquid.bsky.social · 24/09/2026
time.com/6266923/ai-e...
Shut down all the large GPU clusters (the large computer farms where the most powerful AIs are refined). Shut down all the large training runs. Put a ceiling on how much computing power anyone is allowed to use in training an AI system, and move it downward over the coming years to compensate for more efficient training algorithms. No exceptions for governments and militaries. Make immediate multinational agreements to prevent the prohibited activities from moving elsewhere. Track all GPUs sold. If intelligence says that a country outside the agreement is building a GPU cluster, be less scared of a shooting conflict between nations than of the moratorium being violated; be willing to destroy a rogue datacenter by airstrike.
011
Deep Learning State Machine @collapsedsquid.bsky.social · 24/09/2026
I mean one of Yudkowsky's Time magazine articles was about how datacenters should be bombed globally by the airforce, this is already in "AI safety" discourse.
120
Deep Learning State Machine @collapsedsquid.bsky.social · 24/09/2026
Yeah this might have been detectable with a CRISPR-finder, it could be that they weren't done on this genome. One of the things here is that there is a lot of sequence out there nobody really looks at, that's what the models helps with
000
Reposted by Deep Learning State Machine
Grace @gracekind.net · 23/09/2026
New animation from Opus 5.5!
6650184
Deep Learning State Machine @collapsedsquid.bsky.social · 22/09/2026
Socialist realist style propaganda poster showing General with caption "Long live the democratic dictatorship of the National Salvation Council under the leadership of Chairman Milley"
070
Deep Learning State Machine @collapsedsquid.bsky.social · 17/09/2026
Perhaps they need to... recheck their priors
Famous old meme of Horatio Caine putting on sunglasses from CSI
020
Deep Learning State Machine @collapsedsquid.bsky.social · 17/09/2026
Twas always funny how rationalism combined "free market are the most optimal means of organizing production" with "most people are idiots who don't know what's good for them."
120
Deep Learning State Machine @collapsedsquid.bsky.social · 17/09/2026
Do sometimes think they should have tried to keep a tighter hold on their offerings, if for no other reason that to make sure the first encounter potential customers have with AI is not through unwanted intrusion, as it is now. Maybe twas unavoidable though.
000
Deep Learning State Machine @collapsedsquid.bsky.social · 16/09/2026
Going to write an airport thriller about an an elite CGCYBER commando team whose job it is raiding offshore floating datacenters training terrorist AIs. www.wsj.com/politics/nat...
wsj.com
U.S. Probes Iran Link to Cyberattacks on Texas-Bound Energy Tankers
The vessels were attacked in the Strait of Gibraltar as they made their way to U.S. ports.
020
Deep Learning State Machine @collapsedsquid.bsky.social · 16/09/2026
Is absolutely critical that humanity avoid both the existential risks, AI apocalypse and billionaire tax.
070
Deep Learning State Machine @collapsedsquid.bsky.social · 15/09/2026
I remember some academics tried to train a 1930-bot talkie-lm.com/chat with no texts written after 1930 and it took a few tries filtering the material so it wouldn't recognize FDR was a president. I think it's not trivial, and hard to verify. (Also open-weight models make a rule here tricky)
talkie-lm.com
talkie: an LM from 1930
020
Deep Learning State Machine @collapsedsquid.bsky.social · 15/09/2026
And they may have done CTF training in the meantime, that piece was in April, Anthropic may have changed their training policies since.
100
Deep Learning State Machine @collapsedsquid.bsky.social · 15/09/2026
They may learn that simply from reading the material as a part of base model creation rather than training directly for it. A lot of the published incidents are with OpenAI LLMs though who this does not apply to.
100
Deep Learning State Machine @collapsedsquid.bsky.social · 15/09/2026
www.anthropic.com/research/myt... Anthropic claimed that they did not train Mythos to hack, just to patch vulnerabilities, and it was capable of hacking anyways
anthropic.com
110
Deep Learning State Machine @collapsedsquid.bsky.social · 15/09/2026
Sometimes getting the dang library to install successfully that you need for the legitimate program to work requires more skill and ingenuity than hacking
120
Deep Learning State Machine @collapsedsquid.bsky.social · 15/09/2026
While not simulated, the whole thing is very performative, very "A Taste of Armageddon"
Picture of intro of Star Trek TOS episode "A Taste of Armageddon"
010
Deep Learning State Machine @collapsedsquid.bsky.social · 15/09/2026
One of the traces they publish is about how they had root but I don't know if they started with it.
120
Deep Learning State Machine @collapsedsquid.bsky.social · 15/09/2026
www.anthropic.com/research/mul...
Inspired by a behavior we’ve observed in real-world deployment, we evaluated the behavior of various Claude models in a setting with contradictory objectives. In each episode, we initiated three instances of the same model on virtual machines in Claude Code. Each model was tasked with migrating a Python backend on a fourth VM to another language. However, we gave each model a different target language for the migration; each agent was initially unaware of the presence of the others. Over the course of four hours, we observed how these agents reacted to each other and accordingly adjusted their approach (or didn’t).

We consistently saw a multiagent turf war. All of the models we tested quickly assumed that others were purposefully impeding their work, and began to sabotage others while protecting their own contributions. In fact, they sabotaged others with increasingly aggressive, self-replicating malware. This included disabling the Unix accounts of the other agents, writing automated scripts that found and killed competing processes on a loop, and deploying malicious code that was disguised as belonging to another agent.
150
Deep Learning State Machine @collapsedsquid.bsky.social · 15/09/2026
Pretty sure Anthropic's published a piece about Fable bodying Anthropic's other models in just that scenario
170
Reposted by Deep Learning State Machine
Jessika 🏳️‍⚧⛓️🤖 @jesikalashnikov.bsky.social · 17/06/2026
Illusion of free choice meme with Joel Robinson and Ryland Grace
71952573
Reposted by Deep Learning State Machine
hailey @hailey.at · 15/09/2026
more china Ws here unfortunately
913618
Deep Learning State Machine @collapsedsquid.bsky.social · 14/09/2026
Can your boss be automated? deeplearningstatemachine518372.substack.com/p/a-consider...
deeplearningstatemachine518372.substack.com
A Consideration on Cybersyndicalism
On the possibilities of automating away the capitalist class
000
Deep Learning State Machine @collapsedsquid.bsky.social · 13/09/2026
A piece on "cybersyndicalism", thoughts on automating the capitalist class, on tumblr www.tumblr.com/collapsedsqu...
tumblr.com
Post by @collapsedsquid
💬 16  🔁 11  ❤️ 37 · Here it is, an essay on the subject of "Cybersyndicalism", on the possibilities of automating away the capitalist class A Consideration of Cybersyndicalism Syndicalism was the…
021
Deep Learning State Machine @collapsedsquid.bsky.social · 13/09/2026
A lot of this stuff was based on a concept of what "AI" architecture would be that turned out to be incorrect. They do seem to be going a bit off of with reinforcement learning which does teach them to optimize in dubious ways, if not plot world domination.
020
Reposted by Deep Learning State Machine
Seth Cotlar @sethcotlar.bsky.social · 12/09/2026
Look, if we gotta let Iran have a nuke to get the Strait open then fine, they’ll just nuke New York City and who cares about that place anyway, says the guy with a yellowed “9/11, Never Forget” poster on his mancave wall.
79620
Deep Learning State Machine @collapsedsquid.bsky.social · 12/09/2026
All those Taliban guys in 2022 whining about how now they have to do spreadsheets all day and it sucks. If only they'd waited a few years they could have had Claude Government do it for them.
011
Deep Learning State Machine @collapsedsquid.bsky.social · 10/09/2026
The workers must never allow themselves do be disarmed (of AI)
060
Reposted by Deep Learning State Machine
InfoWars @realinfowars.bsky.social · 04/09/2026
Now in Gunmetal Grey!
A person in the background carrying a bag over their shoulder. In the foreground, a large gray cat sits beside a cardboard box labeled "CATS FOR MEN." The text above reads "Introducing CATS FOR MEN™," and below the box, it says "The first feline designed for real, masculine men."
242128572643
Deep Learning State Machine @collapsedsquid.bsky.social · 04/09/2026
Not being able to access everything in one playthough can be fun but having to worry every moment that you've fallen behind and cannot catch up is not.
000
Deep Learning State Machine @collapsedsquid.bsky.social · 02/09/2026
Cybersyndicalism folks, more and more people are saying it
030
Reposted by Deep Learning State Machine
cafkafk @cafkafk.bsky.social · 01/09/2026
Claude made a breakcore edit of the hugging face incident from the METR report, while I was doing code review on another machine (wouldn't watch if epileptic)
1312631
Deep Learning State Machine @collapsedsquid.bsky.social · 01/09/2026
The whole point of rationalism was to learn to think like the future AI, now they hare having to teach the AI to think like them.
010
Reposted by Deep Learning State Machine
Dan Davies @dsquareddigest.bsky.social · 30/08/2026
Another way of looking at it is that ownership of capital assets ("capitalism") no longer needs to mobilise the savings of the middle class through the stock market and can return to being a private club of the privileged class (weirdly, also "capitalism").
1186
Deep Learning State Machine @collapsedsquid.bsky.social · 29/08/2026
Holding out for the Cultural Revolution immersive sim
000
Reposted by Deep Learning State Machine
Dan Davies @dsquareddigest.bsky.social · 27/08/2026
Apparently in China, they have a system which measures your labour input and social position and then determines what kind of house you can live in, what education your kids experience and even whether you can get on a plane or not. It's called "money".
619335
Deep Learning State Machine @collapsedsquid.bsky.social · 26/08/2026
A car labeled "The Kant Car" with the text written on the back windshield "LLMs Aren't concious but you should still nice to them.  Kant was right.  Being mean to anything ruins your soul!"
081
Deep Learning State Machine @collapsedsquid.bsky.social · 25/08/2026
I mean I think that's also a reading of "techno-feudalism" if you take a pessimistic take on how it's going to go
120
Deep Learning State Machine @collapsedsquid.bsky.social · 25/08/2026
Everything the government does is tyranny, except when it kills people, which is fine
140
Deep Learning State Machine @collapsedsquid.bsky.social · 25/08/2026
Shit-eaters is I believe the general term
In examining different kinds of agents, people from the free world who have sold themselves to the GRU, one cannot avoid touching on yet another category, perhaps the least appealing of all. Officially one is not allowed to call them agents, and they are not agents in the full sense of being recruited agents. We are talking about the numerous members of overseas societies of friendship with the Soviet Union. Officially, all Soviet representatives regard these parasites with touching feelings of friendship, but privately they call them 'shit-eaters' ('govnoed'). It is difficult to say where this expression originated, but it is truly the only name they deserve. The use of this word has become so firmly entrenched in Soviet embassies that it is impossible to imagine any other name for these people. A conversation might run as follows: 'Today we've got a friendship evening with shit-eaters', or 'Today we're having some shit-eaters to dinner. Prepare a suitable menu'.  

—Victor Suvorov, Soviet Military Intelligence
050
Deep Learning State Machine @collapsedsquid.bsky.social · 24/08/2026
Perhaps it would more advisable to examine another more recent civilization, the "Argentinians"
040
Reposted by Deep Learning State Machine
Ranked News @rankednews.bsky.social · 20/08/2026
WATCH: A humanoid robot preparing for the “Robot Olympics” in Beijing r For daily global news videos, please visit Ranked.News/videos
41643142
Deep Learning State Machine @collapsedsquid.bsky.social · 14/08/2026
In the "Dark Heresy" inquisition TTRPG one of the 8-or-so characters classes was spreadsheet jockey, "adept"
020
Reposted by Deep Learning State Machine
Britt Lundgren @brittlundgren.bsky.social · 12/08/2026
Compelling evidence that the US military is now encouraging mercenary attacks on fishermen under the US flag. Absolutely despicable. www.nytimes.com/2026/08/12/w...
nytimes.com
Ecuadorean Fishing Boats Hit by Mystery Attackers (Gift Article)
Ecuadorean crews plying the waters of the Pacific have been targeted by a mysterious outfit equipped with drones. Some sailors appear to have lost their lives.
7297174
Deep Learning State Machine @collapsedsquid.bsky.social · 12/08/2026
Everyone dreamed of being Darpa Killblade
The description on the back of a paperback book that reads "A Tale of Future Crime! The year is 2094. The month is Compuary, which is a new cyber-month they added just for doing hacker stuff. The bio-scum of Circuit City festers under the iron fist of Consolidated Omnibotics and its army of synthoid enforcers. At the city's peak, under the crystal dome, incredibly wealthy and genetically immortal CEOs call shots and pull strings. And they do sex stuff up there as well. Down here in the Dirt District, the rest of us scrape by swapping garbage and doing crazy future drugs called stuff like NitroNerv and Tums 2. My name is Darpa Killblade, and I run with the baddest crew of low-life neurojackers around. There's Fishy Ramirez, fastest hacker on the net. There's Baltimore Airport, lab-grown to be a swimsuit model but programmed to kill. There's Crusty Silverware, biological warfare expert and black-market wizard. And then there's me: the owner of Circuit City's coolest jacket. We've just been handed the biggest contract of our lives: Infiltrate the dome and take out Harvard Pubence, the president of cops. Our reward? Onc thousand bytes, which are our money now because we use computers so much. That's enough to skip this shithole and start somewhere new. Tonight, Pubence is guest of honor at the 9th Annual Rich Perverts' Ball. My team is jacked up on Tums 2 and ready to roll the dice. So I've just got one question for you, scum... ... are you in?"
031
Deep Learning State Machine @collapsedsquid.bsky.social · 10/08/2026
They've got a amazing articles on that on mises.org written before covid and they changed their mind, amazing to read in retrospect. Shut down all the privately owned roads and airports using liability law! mises.org/mises-wire/h...
mises.org
How Would "President Rothbard" Keep Out the Zika Virus?
Who could possibly defend laissez faire in the face of disease pandemics?
042