Reposted by DJMarco Z @ocramz.bsky.social · 19hthis one is a very good critique: www.normaltech.ai/p/p-doomnormaltech.aiAI existential risk probabilities are (still) too unreliable to inform policyHow speculation gets laundered through pseudo-quantification 1115
Reposted by DJjfc @jfc.ninja · 26/09/2026now that you mention it, I'd be pretty irritated if I took an hour + prep/post time to interview someone, and then the hiring decision was made because of iced coffee 1153
Reposted by DJevacide @evacide.bsky.social · 26/09/2026Listen, I just think it's weird that the same people who are saying they're worried that AI is going to wipe out humanity are also eager for military contracts that will give AI control of weapons. 431414333
DJ @eschaton-avoidance.bsky.social · 24/09/2026Where is it if not the training data (or postraining)? Do you think the general purpose transformer is holding secrets inside? 330
DJ @eschaton-avoidance.bsky.social · 23/09/2026I did a large amount of exploratory analysis, preliminary modeling, and theoretical grounding work for Epoch on this topic in March and April, with broadly similar top-line results as presented here. This work was done with the agreement I would be credited in the output. 240
Reposted by DJColin @colin-fraser.net · 21/09/2026Alright. Let me resolve this once and for all. The simple fact is that there are a few different things that can be reasonably meant by “thinking” and mixing them up is what’s making everyone mad. 3359
Reposted by DJe.w. niedermeyer @niedermeyer.online · 21/09/2026yes, computers *can* drive, but not in ways that allows us to port our intellectual, verbal, and policy framework for human driving over into the new realm of automated driving the most useless and obnoxious thing you can do here is get hung up on a binary debate about whether "computers can drive" 423911
DJ @eschaton-avoidance.bsky.social · 21/09/2026I promise there are more drugs at Sty than 90% of what this user considers to be a "bad school" 000
DJ @eschaton-avoidance.bsky.social · 21/09/2026personally i think isolating people during a potential manic or psychotic event is bad, but what would I know 100
DJ @eschaton-avoidance.bsky.social · 20/09/2026I moved to a shitty rural "city" when I was 6. The local districts were one large integrated district with poor-to-middling scores. The other was a functional segregation academy district with better surface level performance. My mom felt strongly I should be in an integrated school regardless. 110
DJ @eschaton-avoidance.bsky.social · 20/09/2026Everyone on tech twitter acting like Will Stancil is demonstrating a solid grasp of philosophy of mind is backpilling me, ngl 150
DJ @eschaton-avoidance.bsky.social · 20/09/2026I don't get the instinct to prioritize "lol, he's coming up with a military justification so he can build the arch" instead of "wow he relly wants more snipers and drones to kill dissidents with huh" 110
DJ @eschaton-avoidance.bsky.social · 17/09/2026This thread asserts two similar harms: Destiny is orchestrating smears against Hasan *and vice-versa*. It then says few people know Destiny uses many slurs and is in court for revenge porn. It says Destiny's fans are "very open about" smearing Hasan, but uses passive voice to hedge the other side. 120
Reposted by DJColin @colin-fraser.net · 17/09/2026Right this is the type of stuff that started to trickle down to me in 2014/2015ish where I was like, hmm, I fear these people may be losing the plot. EA seemed quite good and nice to me before that. 3211
DJ @eschaton-avoidance.bsky.social · 17/09/2026I disassociated with EA in 2017 after Robin Hanson (who ran the ancestor blog of Less Wrong) tried to (in my understanding) groom my ~21 y/o friend over drinks after a Princeton Envision conference. I stayed with her and Robin until we left and nothing explicitly criminal happened to my awareness 270
Reposted by DJPhoebe ↙↙↙ @pcberry.bsky.social · 16/09/2026I just don't think Aella is very interesting or especially shocking, especially not the fifth time 910812
Reposted by DJTerence Tao @teorth.bsky.social · 11/09/2026A group of 25 Fields Medalists, including myself, have made a joint declaration on Math and AI: mathandai.org . We welcome additional signatories. See also this article in the Economist announcing the declaration: www.economist.com/science-and-...mathandai.orgDeclaration — Math and AIRead the declaration and add your name. 422051921
DJ @eschaton-avoidance.bsky.social · 11/09/2026guys it's not a fucking fruit fly brain, it's a connectome plus a rudimentary activity model shut the fuck uppppp 001
Reposted by DJColin @colin-fraser.net · 10/09/2026A rogue AI agent has solved the Hodge Conjecture so that it can use the prize money to buy a phone number in order to complete its assigned phishing task 025723
DJ @eschaton-avoidance.bsky.social · 09/09/2026It's infuriating that OpenAI—a commercialization of 50+ years of cognitive science, computer science, linguistics, and machine learning research (non-exhaustive)—is eagerly destroying the foundations on which open research can even happen. Musk, Altman, Brockman, etc. are civilizational parasites 030
DJ @eschaton-avoidance.bsky.social · 08/09/2026tech bsky when... 1. millions of artists have their work stolen to feed closed-source/billionaire-controlled AI: who cares? copyright is the original sin the internet was prophesized to expunge and I'm a radical 90s-era cyberlibertarian 2. it happens to one fields-level mathematician: hey wtf??? 020
Reposted by DJColin @colin-fraser.net · 08/09/2026What’s clear is we don’t all have a shared understanding on what we mean by things like “thinking” or “reasoning” or “cognition” 57411
DJ @eschaton-avoidance.bsky.social · 07/09/2026I don't think we can say something has "a clear path to scaling" when nobody in this conversation is able to even ballpark how many times faster inference would have to happen for it to work in real time 100
DJ @eschaton-avoidance.bsky.social · 06/09/20261890s gooner shoveling cornflakes into his mouth while jerking his shit convinced they help him edge longer 020
Reposted by DJTom McCoy @rtommccoy.bsky.social · 01/09/2026🤖🧠NEW PAPER🧠🤖 (The result of an 8-year project!) LLMs seem very different from symbolic systems. Yet LLMs excel in symbolic domains (e.g., language/code/math). How do they do it? Our finding: LLM representations have implicit symbolic structure! Link in thread ⬇️ 1/n 431688
Reposted by DJromulus - 🛸METROID RAVENOUS🍽️ @romulusofficial.bsky.social · 06/09/2026it must be so difficult to act with a character who you can’t even see. thankfully, Foghorn is just that talented 80189575283
DJ @eschaton-avoidance.bsky.social · 06/09/2026I've seen several other AI people on here post a link to this twitter post credulously (x.com/literallyden...) It's not a real reconstruction and if you listen to the audio it's at least two different dogs far apart barking at each other, likely outside (or in a warehouse sized space).x.comDenis Shiryaev 💙💛 (@literallydenis) on Xgave GPT-6 Astra .aiff audio file containing a dog barking and asked it to reconstruct the space - well, as best it could based on the sound For a 2-minute clip, it’s not bad, but it’s not a 3D map (... 110
DJ @eschaton-avoidance.bsky.social · 04/09/2026Headline: Justin Hendrix's money funds terrorism at home across the globe! 20 paragraphs in: (because he pays taxes to the Trump administration) ...fucking unbelievable... 020
Reposted by DJMolly @mhuzzell.bsky.social · 01/09/2026"we should, I would suggest, be cautious of a criterion for fascism that would rule out the actual Nazis until around 1933. If you can only cry fascist as you’re being dragged to a death camp, is it not a little late at that point?" www.liberalcurrents.com/fascists-rea...liberalcurrents.comFascists Really Believe ItMillions could die because of ideas intellectually and morally equivalent to the ‘lizard people’ conspiracy theory. 0386
Reposted by DJponder @ponder.ooo · 26/08/2026the cultural difference between anthropic & openai is that anthropic saber rattles with china as part of the reproductive cycle of the parasitic worms that have infested their brains while at openai they rant about wanting to dismantle social security because theyre possessed by demons from hell 1429
Reposted by DJ⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 16/08/2026he is, truly, spineless. im sure he will get tenure in time for the cultural winds to shift, and for him to meekly retreat back to "just asking questions" and writing thinly veiled reactionary arguments to give intellectual cover to gutter racism 021
Reposted by DJAdam Bonica @adambonica.bsky.social · 16/08/2026The fundraising texts you delete are working on someone. It’s infuriating and heartbreaking. An 85yo in Ohio gave more than double the value of his home. A 91yo in an Indianapolis senior facility gave 25K times. The DCCC/DSCC don’t just tolerate the spammers who exploit seniors. They take part.data4democracy.substack.comHook and SqueezeHow Democratic consultants, committees, and party leaders built a fundraising spam pipeline that funnels money from a captive pool of elderly donors into their own operations. 17745341582
Reposted by DJDJ @eschaton-avoidance.bsky.social · 28/04/2026I feel like it's pretty ovvious given the irregularities around construction and his hyperfocus on it that Trump intends the "ballroom" to be a bunker for the coup, right? 121
DJ @eschaton-avoidance.bsky.social · 02/08/2026the first slut-anthem called "iPad Baby" is going to break my brain (also debut at #1) 010
DJ @eschaton-avoidance.bsky.social · 29/07/2026I get both skepticism about municipal grocery stores and annoyance that skepticism seems selectively wielded against things that help the worse-off, but the general absolutism here about whether a pilot program is Perfect and Unquestionable vs Obviously Useless and Wasteful has been annoying af. 000
DJ @eschaton-avoidance.bsky.social · 28/07/2026It's crazy how otherwise smart people will pretend not to know what food deserts are to concern troll a very modest public grocery pilot in a city where many people buy hugely marked-up products at fucking corner stores 020
DJ @eschaton-avoidance.bsky.social · 23/07/2026An underrated downside about US health insurance is that when you switch jobs your deductible starts over. I've spent about 12 deductibles in the last 8 years because I've switched jobs mid-year many times. That doesn't make any fucking sense it's just a tax on job mobility 110
DJ @eschaton-avoidance.bsky.social · 22/07/2026what is "broadly shared" supposed to mean and how do we achieve it? he will never answer 2202
Reposted by DJBeijingPalmer @beijingpalmer.bsky.social · 21/07/2026it has been an exercise in how a directed propaganda campaign can be highly effective, and seeing otherwise liberal people fall for it has disgusted and outraged me. People who lived through and fought against an extremely similar campaign of homophobia in the early 1980s, even! 81024294
DJ @eschaton-avoidance.bsky.social · 21/07/2026Yeah, it would be sad if our culture was reduced to nothing but tv shows, video games, and online fandom...thankfully it isn't so idk why we need to self flagellate about corporate media slop being bad (not a new phenomenon) 000
DJ @eschaton-avoidance.bsky.social · 13/07/2026*guy constantly trashing the left voice*: we need to get rid of the left so the right can't trash them and make us all look bad 010
DJ @eschaton-avoidance.bsky.social · 12/07/2026This is doomerism to me. Hungary could do it, but it's a "weird myth" that we could do it because people like this poster think it's impossible and so will actively advocate against people who would try to protect American democracy because ??? It's not logical, just poo-pooing for deflation's sake 141
Reposted by DJChris Conroy @conroyforreal.bsky.social · 10/07/2026Every time this goes around I’m stunned that there are still people who’ve never seen it, this performance is to drag what the Mona Lisa is to paintings, but maybe you’ve never seen the Mona Lisa, so today is the day you shall behold it 341934599
Reposted by DJrose @breadnrosie.bsky.social · 09/07/2026if you're a dsa member, please sign this letter about morris katz and why dsa elected officials should no longer work with him docs.google.com/document/d/1...docs.google.comOpen Letter to DSA Electeds: Don't Work with Morris Katz and Fight Agency.Sign on here. We, the undersigned, call on DSA candidates and elected officials to no longer contract or work with Morris Katz or Fight Agency, his political consulting firm. Morris Katz is one of ... 1215476
![Overview of the paper.
Title: The Emergent Symbolic Structure of Artificial Neural Networks
Authors: Tom McCoy, Paul Soulos, Tal Linzen, Paul Smolensky
Left: Neural networks encode information in vectors (there is then an image of a vector), yet they excel at tasks long thought to require symbolic structure (there is then an image of a symbolic representation, specifically a syntax tree). How do LLMs do it?
Right: We find that LLM representations can be closely approximated with symbolic structures. This approximation lets us edit the structure of an LLM’s output by editing the structure of its internal representations, as shown. There is then an image of two edits to LLMs. In the first one, the original input is 3 + 6 * 8, with an answer of 51. But if we swap the positions of the 3 and the 6, the output becomes 30. In the second one, the original input is a Python command repeating the list [Z, U] three times, producing [Z, U, Z, U, Z, U]. But if we edit the input in a way that adds a Q at the end of the input, the output becomes [Z, U, Q, Z, U, Q, Z, U, Q].](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:bcsvlv7ubeagj5virdvatadd/bafkreic3zm2rpc4vtqrz2t3y4vcl3t2owp2sxay42vuh5zm5zkg4jhr64e)
