O Who's Who da IA

O que a comunidade de IA realmente lê e debate: os artigos que seus pesquisadores, criadores, especialistas em ética e críticos compartilham, e as discussões que eles têm, classificados pelo peso dentro da área.

O que o diretório está lendo

O que eles estão debatendo

7 membros · 28 posts ·6d ago
Ted Underwood · Ciara Rowles · Grace · tweety fish · mackuba.eu · Brett Bobley · Emile van Krieken · Luis Villa
Ted Underwood: This divide is also why half of my friends are "What will we put in the abandoned shells of these data centers when people realize they're useless?" and the other half are "Can …
Ted Underwood: Meanwhile memory chips are increasing in price so fast that by the time Neuromancer comes out this fall you will *actually* be able to get killed in Chiba City for 3 MB of hot RAM.
Ted Underwood: The tacit premise there is that coding is a narrow, purely technical application with an obvious ceiling on demand. (Which is what half of my friends do assume.) But I'm startin…
Ted Underwood: I don't belong to an evangelical religion, so that sounds good to me. Live a code-free life; enjoy summer; plant tomatoes, they're delicious!
Abrir a thread →
7 membros · 29 posts ·14d ago
Ted Underwood · Brian C. Keegan, Ph.D. · Derek Willis · Thomas Dietterich · Marco Z · Mark J. Nelson · Matthew Kirschenbaum · Hagen Blix
Ted Underwood: Every academic I know who uses or studies AI is also deeply worried about the technology’s effect on universities. Concern is universal. What separates people is how they think …
Ted Underwood: People trying to steer through this typically see enough value in the technology to believe braking isn’t likely to work. (Because adoption in & outside universities will not sl…
Ted Underwood: Anyway, increasingly what I want to hear is not what worries people — we’re all worried about the same things — but what people believe should be *done* medium-term, and especia…
Brian C. Keegan, Ph.D.: I think the academy should spent a few decades litigating which metaphor is the most apt for theorizing, which disciplines are allowed to have opinions, and subjecting our stude…
Abrir a thread →
6 membros · 14 posts ·13d ago
Eugene Vinitsky · Ted Underwood · Maxim Raginsky · Ida Momennejad · Fernando Diaz · Ethan
Eugene Vinitsky: arxiv paper now on hold for almost a week. To all the people submitting LLM slop that brought us to this place, please know that you've appreciably made the world worse
Eugene Vinitsky: No issue with using an LLM in the loop, but tremendous issues in submitting work that you can't stand by and causes challenges for everyone downstream of you
Eugene Vinitsky: I think all my LLM complaining should really be more interpreted as complaining about people. You don't have to do this! LLMs don't submit slop, humans submit slop
Ted Underwood: I've heard waits of a month and more, at which point it becomes more like a postprint server.
Abrir a thread →
7 membros · 20 posts ·15d ago
Tim Kellogg · austin · Chris Paxton · Naomi Saphra · Mark Riedl · Joanna Bryson · Phillip Carter
Tim Kellogg: Opus and GPT-5.5 are both below 2T, DeepSeek can catch up to both, Fable will keep increasing the gap
austin: “meta and xAI are still planning” what, their irrelevance?
Tim Kellogg: they’re thinking about doing something, maybe
Tim Kellogg: i think so, but someone will surely pushback
Abrir a thread →
6 membros · 31 posts ·15d ago
SE Gyges · tweety fish · Eli Sennesh · Maxim Raginsky · Grace · Singularity's Bounty e/cc
SE Gyges: lmao holy shit are they actually doing yud's plan? this is yud's plan
SE Gyges: i think it is a stupid plan and have mocked it in public
tweety fish: Yud's magic is that he's too far up his own ass to know that he's swindling dumb rich people
SE Gyges: i don't even care, you're not getting useful work out of your expensive ubermenschen even if you can raise them. we already have billions of people, supply is not the bottleneck
Abrir a thread →
6 membros · 22 posts ·19d ago
SE Gyges · critical slop studies · Isaiah Bishop · jake · Liz Fong-Jones (方禮真) · Eli Sennesh
SE Gyges: so far as i can tell everyone in san francisco who does anything in ai policy is part of one large polycule
critical slop studies: further overdetermination re my never wanting to take a job in SF
SE Gyges: no, actually, i am just catching people's sloppy drama on the internet. i am in seattle and i know who's fucking who at manifest
SE Gyges: at least there's one!
Abrir a thread →
4 membros · 11 posts ·5d ago
Ethan Mollick · Eli Sennesh · Singularity's Bounty e/cc · Fran Litterio · Ted Underwood
Ethan Mollick: It is weird that there is still a substantial set of people who believe "AI is mostly hype" at this stage: Five Eyes is warning about AI, exponential revenue & token use at the …
Ethan Mollick: I know I write a post like this every month, but the degree to which the refusal to believe that maybe this is not just a fake bubble thing is getting weirder all the time. And …
Ethan Mollick: And if AI is real, than even if there is a financial bubble, it just does not matter. In a world when OpenAI and Anthropic collapse under pressure, their IP will just be picked …
Ethan Mollick: You can believe there is a financial bubble if you would like. I can't guarantee there isn't one, as I discuss in the thread. It doesn't really matter much. And if you don't tru…
Abrir a thread →
6 membros · 9 posts ·12d ago
Ted Underwood · SE Gyges · Singularity's Bounty e/cc · Cameron · jake · Luis Villa
Ted Underwood: Is anything creepier than a language model flattering you using its knowledge of previous convos? ("As you know," or "this is where your prior work on Y will really pay off") A …
Ted Underwood: Yes, I know I could just turn off memory, but ... it's a really smart stalker ghost, and sometimes it's right? I mean, clearly it's a good judge of character.
SE Gyges: it cues incorrectly from memory way more often than correctly for me. i had to grudgingly admit that the basic .md file was a good idea
Ted Underwood: yeah, this is a pretty common pattern *its* understanding of what I was doing is quite different from *my* understanding
Abrir a thread →
3 membros · 7 posts ·18d ago
Eugene Vinitsky · Christopher Mims · Ishan Khatri
Eugene Vinitsky: To understand what's happening, you need to understand that some people genuinely, without any artifice, believe we are getting recursive self-improving intelligence in the next…
Eugene Vinitsky: I'm not commenting on the validity of those claims. But this is a real position people are operating on and if you don't understand that folks with power take this seriously a l…
Eugene Vinitsky: If you're going in with "they're all faking it to make money" you're going to be very confused by why things are happening as they are
Eugene Vinitsky: I think a significant part of the administration does believe in the "we must lock in the fascist empire forever through ownership of AGI"
Abrir a thread →
6 membros · 19 posts ·18d ago
SE Gyges · Adverb · jake · mackuba.eu · Isaiah Bishop · Singularity's Bounty e/cc
SE Gyges: it is currently illegal for amanda askell to talk to the new claude, and if this stands anthropic is probably dead
Adverb: I think there is enough money in these IPOs that Anthropic will come out on top
SE Gyges: check my profile and scroll back
SE Gyges: scottish
Abrir a thread →
6 membros · 8 posts ·18d ago
Sung Kim · Gus · Dustin Moskovitz · austin · Singularity's Bounty e/cc · Ramon Astudillo · Chris Paxton
Sung Kim: Just FYI on Anthropic's Fable 5 fiasco.
Gus: I don't know. It's very hard to believe them if you look at their track record
Dustin Moskovitz: fix the jailbreaks smdh alignment was staring us in the face the whole time!
austin: mythos, fix the jailbreaks in fable. make no mistakes.
Abrir a thread →
5 membros · 19 posts ·3d ago
Ted Underwood · jake · Tim Kellogg · Max Woolf · apenwarr.ca
Ted Underwood: Lately I hear academic friends celebrate open-weight or local models for a mix of three reasons: a) transparency is good for research b) they protect higher ed's autonomy from S…
Ted Underwood: The same tasks may actually use less ⚡️ in a center, b/c better hardware & caching. “I use smaller models locally”—maybe? but they cost less, so Jevons paradox can kick in. Micr…
Ted Underwood: Jevons paradox is a problem even for "smaller." I run tens of thousands of queries on small models, because it's effectively free; I think twice before doing that on frontier mo…
Ted Underwood: The cost of paying frontier labs is absolutely an incentive to go local. But I'm skeptical that there's much incentive to be efficient—because, in reality, these things just don…
Abrir a thread →