Jamil Civitarese @jamilkpc

Responding to the four IDF soldiers who were reportedly killed by Hezbollah, Itamar Ben-Gvir, Israel’s national security minister, uses unquestionably genocidal rhetoric in this post. This man is a danger to the security of Israel and to global Jewry. If there were any justice, the United States would use every tool in its diplomatic and legal toolkit to punish him, including directly sanctioning him, barring him from entering the United States, restricting his access to all U.S. information, and cutting off all nonessential contact with his ministry. I’m not suggesting this has a chance of happening, nor am I suggesting he cares or would change, but it must be done. No person who speaks this way should be treated as a legitimate partner by the United States in any respect. And no Israeli official who calls for an entire country to burn should be allowed to hide behind the language of security while making Jews around the world less safe.

"Transformers" by Daniel Jurafsky and James H. Martin is one of the clearest and most mathematically grounded introductions to the Transformer architecture I have ever read. Chapter 8 introduces the Transformer as the standard architecture behind modern large language models. What makes this chapter particularly interesting is its step-by-step presentation of the underlying mechanisms: contextual embeddings, self-attention, query, key and value vectors, scaled dot-product attention, multi-head attention, residual streams, feedforward layers, layer normalization, masking, and the parallel matrix formulation of attention. In particular, the treatment of attention as a weighted sum of contextual representations is especially valuable. The chapter first develops an intuitive, simplified view of attention and then gradually derives the full formulation using the Q, K, and V matrices. This approach makes it easier to understand what is actually happening inside the architecture from an algebraic and matrix-based perspective, rather than simply viewing the usual block diagrams. I think it is an excellent resource for anyone interested in understanding how Transformers work from linguistic, mathematical, and computational perspectives. web.stanford.edu/~jurafsky/slp3…

Não sei se é uma má ideia. Muito da produção científica feita ou intensamente assistida por AI não é tecnicamente errada, mas falha esteticamente no que cientistas consideram bom. O problema é que isso aí tem tudo para criar uma FubangAI, gosto em design é muito sutil.

Thais Castello Branco @thaiscbranco_

3 days ago

We’re excited to introduce Taste Labs. Our mission is to end AI slop. We’re building the data and infrastructure layer to give AI models and agents taste. And today we’re coming out of stealth, announcing our $18.5M seed funding, co-led by @CRV and @AmplifyPartners AI has

585 203 4K 2.5M 3K

New paper with @NAChristakis in @ASR_Journal: Targeting well-connected people for behavioral interventions is standard practice. We argue it doesn't always work. When ties concentrate around those individuals, this network structure reduces diffusion. doi.org/10.1177/000312…

Uma parada que muita gente não entende sobre IA é que não tem como competir em modelos de ponta. Só o que a Meta - que está bem pra trás - está gastando em data centers este ano é em torno de 5% do PIB do Brasil. É bonito gritar “soberaniaaaaa” mas sem entender isso só vai levar a ser iludido por aventureiros. Isso não é nem algo que se aplica só pro Brasil. A Europa está num dilema pq eles simplesmente não têm dinheiro suficiente pra competir em IA de ponta. Pra fins de comparação, a Meta vai gastar em data centers este ano +- que a Alemanha gasta em defesa nacional. O que dá pra fazer então? O que dá pra fazer é se tornar parte da cadeia produtiva. Criar condições atraentes pra instalação de data centers, expandir produção de energia pra oferecer energia barata (é bom pros cidadãos também inclusive) e se manter integrado internacionalmente. De certa forma isso torna os países que fizerem “acionistas” da IA. É o jeito de garantir uma fatia.

Durante um jogo do Brasil, comércio fecha, rua esvazia, bar enche e nós associamos a performance da seleção ao valor da nossa nação Pra mim, NADA do que se envergonhar: o Brasil ama futebol. Copa não precisa ser "pão e circo". Pode muito bem ser a união que precisamos, aquela brisa de alento No mais, a seleção tem OBRIGAÇÃO de dar tudo de si, é só isso que o brasileiro pede e a gente merece Além de todas as necessidades materiais, o que o BR mais precisa são muitos tipos de PAZ. Um hexa ajudaria

Crows (especially American crows) are highly intelligent and have remarkable abilities to recognize individual human faces, remember negative experiences with specific people for years (up to 17 in some studies), and socially transmit that information to their offspring and other crows, even those who never directly experienced the original incident. Researchers wore distinct masks while trapping and banding a small number of crows (a stressful but non-harmful experience for the birds). The crows quickly learned to associate that specific "dangerous" face with threat and responded by scolding, mobbing, and dive-bombing anyone wearing it. Offspring and other crows picked this up through observation: young birds (fledglings born later) would join in or react independently after seeing their parents' alarm calls and behavior. This "knowledge" spread through the local crow population via social learning, turning a personal grudge into a community warning. [Cornell, H. N., Marzluff, J. M., & Pecoraro, S. (2012). Social learning spreads knowledge about dangerous humans among American crows. Proceedings of the Royal Society B: Biological Sciences, 279(1728), 499–508. DOI: 10.1098/rspb.2011.0957]

@Leukonomics Acho normal pra uma seleção que ainda não se encontrou muito. Eu daria mais chances, mas você tem um ponto

@Leukonomics Achei Paquetá ok dentro do possível.

A GENTE VAI NORMALIZAR MESMO O JOGADOR RAPHINHA SOLTO NA SOCIEDADE????

@Erscheinuung @luizcr89 @Leukonomics Caralho, até eu tô fazendo o L agora

A wild scenario. I do not agree with much of the economics. But the key is correct and part of an op-ed I have coming out shortly: very soon, the Q for non-US/China will be whether they *are allowed* to use models from those countries, not whether they can *avoid* doing so. 1/4

Judith Dada @DadaJudith

a week ago

Most of Europe has not yet absorbed what AI is about to do to us. The few who have are not saying it loudly enough. We wrote Europe 2031: a five-year scenario of the continent's slide into irrelevance, how AI is driving it, and what can still be done to change course.

93 171 942 376K 596

É HOJE!!!!

Increasingly coming to terms with the fact that AI was never meant to be given to the masses, it's very clearly a good thing for 5% of the population and a cognitohazard for the rest

The Washington Times @WashTimes

2 weeks ago

EDITORS' PICK | The statistical impossibility of L.A.'s mayor race trib.al/jNOxJg4

112 79 267 1.2M 82

O valor da torre de marfim é evitar que a melted videos faça a gente discutir algo superado há décadas: é consenso, é documentado, já sabemos tudo que precisamos sobre MCT e política econômica no Brasil. Não faz sentido debater isso sem fontes novas.

Guilherme Vianna @DonVianna

a week ago

3 121 889 20K 18

I'm delighted that this short survey, a labor of love, is forthcoming in Notices of the AMS later this month. I'll post a few threads over the next few weeks. Today: why do eigenvectors keep showing up in models of networks and influence?! 1/

Great to see the new generation of political economy coming out from Brazil —- Elections that inspire: Effects of Black mayors on educational attainm... sciencedirect.com/science/articl…

I’ve been meaning to write a tweet thread about a new paper— Barriers to Adopting Predictive Algorithms: A Criminal Justice Field Experiment. We built fancy ML sentence prediction software for public defenders and they mostly didn’t use it. We then pivoted to asking why not. 1/

Arpit Gupta @arpitrage

2 weeks ago

Things look good for AI as a “normal technology” so far. LLMs have rapidly impacted coding, but few other professions; and even in coding it’s hard to see value generation. Bottlenecks, regulation, other frictions keeping humans complementary and relevant

26 17 149 39K 23

Is it the phones? A new paper uses AT&T's 2007–2011 iPhone monopoly as a natural experiment and finds that iPhone access sharply reduced births, especially among younger women. Smartphones may explain 33–52% of the post-2007 decline in U.S. fertility.

Important from a great team of labor economists. I am open to the benefits of remote for certain workers, but as with masks in public, my starting point is "societal interaction among strangers is a key part of humanity and it is bad that we lost so much of it from 2020."

Natalia Emanuel @NataliaHEmanuel

2 weeks ago

Remote work hurts mental health. New research out in Science with @emma_k_h and Amanda Pallais. science.org/doi/10.1126/sc…

4 48 203 99K 116