Confirmo que há todos os indícios típicos de bloqueio nacional determinado pela Anatel no dia de hoje para api.github.com. Nas quartas-feiras, geralmente em dias de jogos de futebol, a agência de reúne com as maiores operadoras do país e determina o bloqueio de endereços utilizados pelos TV boxes. A lista de endereços bloqueados é mantido em sigilo pela agência, algo que tenho criticado em artigos, entrevistas e palestras. Os indícios são: 1) O fato de ocorrer numa quarta-feira e pelo relato desse internauta que me acionou, ontem o problema não ocorria. 2) O fato do IP 4.228.31.149 para qual aponta o FQDN api.github.com estar bloqueado somente na Claro, Vivo, Nio, Algar e TIM e *NÃO* em ISPs regionais conforme posso testar com o comando globalping. 3) O fato de outros endereços IP contidos no mesmo bloco /24 serem normalmente alcançáveis, o que exclui problemas de roteamento já que todos os IPs entre 4.228.31.1 e 4.228.31.255 necessariamente pertencem a mesma rota (vide exemplo 4.228.31.3). Recomendo que os afetados entrem em contato com seus provedores mostrando evidências coletadas a partir de sua casa ou empresa e exijam uma resposta do porquê não havia rota para 4.228.31.149 na noite de 10/06/26 enquanto para 4.228.31.3 há. Nesse thread a seguir, colocarei algumas informações úteis sobre o problema.

The UEFI firmware on motherboards from ASUS, Gigabyte, MSI, and ASRock tells the operating system the IOMMU is ready for DMA protection during early boot. In reality, the IOMMU initialization fails silently on a wide range of these motherboards. A malicious PCIe device with physical access gets unfiltered memory access before the OS loads any defenses. The firmware asserted the lock was engaged.

I can see the seething in the comments. The issue with ARM platforms is they lack an equivalent of BOIS/UEFI. The real magic of the PC is the BIOS. That allows for a far more open platform than ARM or even RISC-V. If you care about open computing, it's x86 all the way

Tô num server de ML no Discord Decidiram usar MÉTODO DE NEWTON numa rede neural, a justificativa dos caras foi: “Vai convergir em menos épocas” kkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkk demorou 7 minutos para a rede treinar usando o dataset WEB10K

Imagine if codex existed in 2007 (last one)

Young people just getting into vulnerability research and exploit development today days have it soooo much harder than those who started even as little as 10 years ago. Binary exploitation is at least an order of magnitude more difficult today than in 2015. Especially if you have zero experiencing doing it. For those who are experienced, the difficulty levels have risen *somewhat* gradually. Though many still haven't been able to keep up. And even among those who could keep up, there are many who have just found it to be so hard as to no longer be fun, and have moved on to other things. Some have posited that binary exploitation has become so hard that the future will be logical/functionality abuse. And while this category of bugs certainly shows a quite a bit of promise in second order exploitation components like LPE, TCC bypass, etc—memory corruption still reigns supreme when it comes to initial access. Memory corruption that is actually exploitable in a reliable, near-instant and deterministic manner—versus random non-useful null ptr dereferences in browsers that require spraying whatevers to get 60% reliability—has gotten so incredibly tough (at least in the most contested systems) that there may be as few as 400 or 500 people in the world today that can actually pull it off repeatably. Seriously, it's probably 400 or 500 people (just my best guess) And that's *with* the help of AI, custom/expensive/gov-only tooling, insider insights/experience and any other resources imaginable. So if you don't make it into the elite of binary exploitation, don't beat yourself up. It's a group of people roughly as rare as NBA players. However, unlike the NBA, it's not obvious who is 7 feet tall with a 40 inch vertical, and who isn't. So it doesn't hurt to try—as long as you're having fun.

Three years ago, when CS2 was just a dream, this man asked for only three things: - 128-tick Servers - Source 2 Engine - Better matchmaking Do we actually have any of it? 🤔