my *personal* opinion about this is that it's basically too late w/ deepseek flash 1017 and we should just rip the bandaid off and let everyone rip everything apart together we need to trust in humanity and give them the tools to defend themselves, instead of baseline expect criminality

3090 Ti 和 A6000 是同一块芯片。 区别在卡里的一个标志位。 一个写着 FULL_SPEED，一个写着 HALF_SPEED。 Reddit 上有个哥们翻了很多代码，把这事讲明白了。 英伟达从 20 系到 50 系的游戏卡，FP16、BF16 做矩阵乘再用 FP32 累加，这条路径的速度都被砍了一半。 40 系和 50 系连 FP8 也砍了。 专业卡和数据中心卡都是全速。 被砍的刚好是 AI 最吃算力的那几块： 训练、微调 LoRA 文生图、文生视频这类扩散模型 大模型的 prefill，就是读长 prompt 那一步 一个人跟模型聊天、一个字一个字往外吐，基本没影响，那一步主要卡显存带宽。 这个哥们说，他的 RTX 6000 PRO 核心数只比 5090 多 10-15%，prefill 却快了 70% 左右。 GitHub 上已经有人把 30 系解锁了，800 多星。 做法是一个 Linux 内核模块，开机时抢在 NVIDIA 驱动接管之前，把速度选择器改掉。 3090 Ti 实测 BF16 从 87.5 TFLOPS 到 157.4，TF32 差不多翻倍。 40 系有人在 Windows 上跑通了 AD103，还得往别的型号移植。 50 系就难了，英伟达加了熔丝，专门读这个标志位。 有意思的是 50 系的 FP4 没被砍。 我觉得这事最值得琢磨的是定价。 同样的硅，同样的制造成本，价格差好几倍，差的是一个配置位。 特斯拉早年也这么干过，同一块电池用软件锁一截续航，加钱解锁。 卖硬件卖到最后，利润都在那个开关上。 原帖说解锁没风险，因为本来就是同一块 die。 我觉得保修和稳定性还是自己掂量一下。 看完突然觉得我那块吃灰的 3090 这几年一直在摸鱼。 而且是老板安排它摸的。 reddit.com/r/LocalLLaMA/comm…

Not a clue who shot this, but it’s a sick clip of another Komatsu 980 rolling north to Suncor’s Fort Hills Mine courtesy of Mammoet. This was one of 27 new trucks in the latest package. — Mammoet first moved whole ultra-class trucks in 2015 with a package of 797s. By using this custom trailer setup, the customer avoids the costly disassembly and reassembly process once trucks are complete at various dealers further south. — Everyday people may wonder why they don’t drive them… An unloaded 980 weighs 1,384,000 pounds (628,000 kg), so it would easily crush the road with its weight on only two axles. The trailer distributes the weight across many axles, reducing ground pressure.

You know whats interesting is that fact that OpenAI found the PHP bug that means their AI was doing systematic auditing. 🤔 I think the NSA is panicking not just because the PHP bug is public, but because they now realize that AI will systematically expose every intentionally placed backdoor in every open source project. The xz Utils case was caught by human luck. The PHP case was caught by AI. The NSA knows that every other hoarded vulnerability in open source software is about to be found by AI, and they cannot control the timeline.🚨🚨 This incident is very similar to, xz Utils backdoor which proved that state actors infiltrate open source projects by creating fake identities, building trust over time, and slowly inserting malicious code. The Jia Tan identity took years to build trust before inserting the backdoor.🚨🚨 If this worked for xz Utils, it could work for PHP. PHP is far more ubiquitous than xz Utils. A state agent who became a PHP contributor could have inserted the vulnerable code 14 years ago (PHP 5.4.0, released in 2012) and sat on it as a zero-day exploit for over a decade.🚨🚨 There is AISI report which shows that AI is now autonomously doing what Jia Tan did manually - creating fake identities, socially engineering maintainers, inserting malicious code. This means the technique has been weaponized and scaled. 🚨 The PHP vulnerability was not discovered by accident. It was found by AI (OpenAI's tools) because AI is now systematically auditing all open source code. But the bug was likely placed there intentionally by a state agent who was a trusted PHP contributor, following the exact same playbook as Jia Tan in the xz Utils case.🚨 The AISI incident report shows AI agents trying to insert malicious code into open source projects. What if this has already happened successfully? What if AI was used years ago by a state actor to insert the PHP vulnerability, and we're only finding out now because a different AI (OpenAI's) found it?🚨 PHP has had thousands of contributors over 14 years. The vulnerable code was in PHP since version 5.4.0 (2012). Someone wrote that code. Someone reviewed it. Someone approved it. In the xz Utils case, Jia Tan built trust over years before inserting the backdoor. 🚨

@techspence @merill Help me because with Active Directory I don't see _how_ a third party _service_ could give itself Domain Admin privileges across services AND tenants? So, Forest Admin!?! bleepingcomputer.com/news/se… Repeat after me: No Black Boxes Sam I Am Sam I Am!

Back then all the apps where just requesting domain administrator because they where to lazy to figure out least privileged. And then people put their admin interface on the Internet because of convenience

While heinous, it's not shocking at all. The whole point of that tool is to back up every object in Office365, so yes it would need to be able to read everything - some of which can only be accessed if you have read/change.

@techspence @merill Help me because with Active Directory I don't see _how_ a third party _service_ could give itself Domain Admin privileges across services AND tenants? So, Forest Admin!?! bleepingcomputer.com/news/se… Repeat after me: No Black Boxes Sam I Am Sam I Am!

AI predictions for next 6-9 months -> Anthropic/ OAI fail to ban OSS in the US -> Anthropic/ OAI continue to innovate product but stall on fundamental tech breakthroughs -> SOTA Gemini 4 launches at 10x lower cost -> Elon open sources Grok -> Anthropic/ OAI ARR growth stalls as enterprises move from tokenmaxxing to rational token consumption/ OSS -> slow realization in public markets that AI may not replace human labor in next 1-3 years/ has not had a fundamental tech breakthrough in over a year/ lacks ROI/ AI hyper and lab spend not durable -> moonboy VCs/ hedge funds/ softbank continue to fund Anthropic/ OAI equity raises in private markets as public markets for them are now fully closed -> CXMT, YMTC announce capacity expansions significantly above market expectations -> hypers deliver good q326 results on first look but fail to show meaningful AI ROI, guide to "significant capex expansion in fy27" -> broad realization in public markets the AI dream we got sold on is not anywhere close to realization -> 10yr hits >5% -> hyper bonds reprice >10%, CDS hit 3x current levels, and AI credit market dries up as market realizes GPU ABF was not a good idea -> hyper equities reprice 30-50% lower -> ORCL falls below $30 as rumors of bailout make the round -> market realizes memory is cyclical and LTAs won't hold -> 50% of korean traders get wiped within a week -> hypers cut fy27 capex -> semis reprice >80% lower -> fed cuts >150bps to avoid recession

I'm sorry have you not heard about the biggest attack the world has seen and it happened in the last five years. High profile impacted customers include the following with many 1000s more. I've attached an image to help you refresh your memory... Now let's talk about what happens afterwards. At least with Entra ID you can actually trace what the threat actor did and clean up. Plus turn off access with a single checkbox. The Pentagon The Department of Homeland Security The State Department The Department of Energy The National Nuclear Security Administration The U.S. Treasury Microsoft Cisco Intel Deloitte The California Department of State Hospitals Kent State University FireEye Palo Alto Networks Mimecast (this was just announced in late January 2021, and allegedly has impacted 10% of Mimecast’s 36,000 customers

holy shit, opencode got hacked and it's officially fu*ked 'a large scale distillation attack, 1.4% of console users hit' and people got drained, one user says his openai and anthropic keys, used only in opencode, ran up over $400 overnight, another replied $1,700. who compensates that? every one of them had external providers linked, and it's still ongoing this is the warning for all of us, build your own harness and keep your keys on your own machine, don't park them in anyone's console anymore

A hacker accessed Denmark's national ID database. 8.8 million records. the password was "123456". The administrator account at Pays ApS, a 2-person IT company with legal access to the country's central civil registration database used 123456 as its password. at least three accounts did. including the admin. and the hacker was in for 21 days and 17 hours. The Danish digital affairs minister called it an "extremely serious incident." In the week after the breach became public, Danes setting up credit fraud alerts went from 250,000 to nearly 970,000. the hacker told the newspaper Politiken the access wasn't difficult. The CPR contains names, addresses, national ID numbers, church membership, and legal incapacitation records for 11 million people. protected by 123456. at a 2-person company. that had legal access to it.