Wouldiwas Shookspeared

@heynavtoor Your entire post is written by an LLM, tho.

🚨BREAKING: Every book you have ever read. Every novel that has ever been published. It is sitting inside ChatGPT right now. Word for word. Up to 90% of it. And OpenAI told a judge that was impossible. Researchers at Stony Brook University and Columbia Law School just proved it. They fine tuned GPT-4o, Gemini 2.5 Pro, and DeepSeek V3.1 on a simple task: expand a plot summary into full text. A normal use case. The kind of thing a writing assistant is built for. No hacking. No jailbreaking. No tricks. The models started reciting copyrighted books from memory. Not paraphrasing. Not summarizing. Entire pages reproduced verbatim. Single unbroken spans exceeding 460 words. Up to 85 to 90% of entire copyrighted novels. Word for word. Then it got worse. The researchers fine tuned the models on the works of only one author. Haruki Murakami. Just his novels. Nothing else. It unlocked verbatim recall of books from over 30 completely unrelated authors. One author's books opened the vault to everyone else's. The memorization was already inside the model the whole time. The fine tuning just removed the lock. Your book might be in there right now. You would never know it unless someone looked. Every safety measure the companies rely on failed. RLHF failed. System prompts failed. Output filters failed. The exact protections these companies cite in courtroom defenses did not stop a single page from being extracted. Then the researchers compared the three models. GPT-4o. Gemini. DeepSeek. Three different companies. Three different countries. They all memorized the same books in the same regions. The correlation was 0.90 or higher. That means they all trained on the same stolen data. The paper names the sources directly: LibGen and Books3. Over 190,000 copyrighted books obtained from pirated websites. Right now, authors and publishers have dozens of active lawsuits against OpenAI, Anthropic, Google, and Meta. These companies have argued in court that their models learn patterns. Not copies. That no book is stored inside the weights. This paper says that is a lie. The books are still inside. And researchers just pulled them out.

@eyenix17 @dangitman50 @AMD @engadget No, it's Zen microarch.

@dangitman50 @AMD @engadget Threadripper is a ryzen......

“This is the company’s most powerful CPU.” @Engadget spotlights the new Ryzen 9950X3D2, packing an incredible 208MB of on-chip cache to push gaming, creative, and AI workloads even further. Learn more: engadget.com/computing/amds…

@ZoeysPosts @AMD @engadget Your AI failed to notice 208MB wasn't even the size of a single memory chip on a typical 2015 PC. The normal amount for a gaming PC was 16GB (16.000MB for you)

We’ve been conditioned to think “more cores = faster.” But AMD just dropped 208MB of cache on a consumer CPU , more L3 than the total system RAM in a high‑end PC from 2015. The uncomfortable truth: For gaming and latency‑sensitive AI, cache is the new cores. The 5800X3D proved it. A chip with fewer cores beat CPUs with twice the core count in games, purely because it had a fat stack of L3. Now AMD is doubling down: stacking cache vertically, using TSMC’s advanced packaging, turning a CPU into a hybrid beast that doesn’t need a separate NPU to handle AI workloads. But here’s the real story: This is brute‑force hardware solving a problem that software should have solved years ago. We’re throwing cache at the wall because memory management, compiler optimization, and game engine data locality haven’t kept up. The 208MB number is a trophy. It’s also a confession: software isn’t getting smarter fast enough, so we’re paying for bigger caches to compensate.

For those who didn’t believe. It was always a long-term plan.

@elonmusk Isnit faster than Roadster, though?

Cybertruck is quicker than a Porsche 911, more powerful than a Ford Raptor

@luisda021298 @johnyesmeralda @mikydressy @fabrimoreg @lauty1990 @Eltomex56 A 4k no importa tanto

@johnyesmeralda @mikydressy @fabrimoreg @lauty1990 @Eltomex56 A una 5080 mínimo le debes poner un proce de serie 7000 o de serie 14 de Intel.

Filtraron las RTX 6000 -Duplica rendimiento con Raytraicing -Mejora un 30% de rendimiento en rasterizado -RTX 6070 y 6080 con 16GB y 20GB -RTX 6090 con 32GB Con esa cantidad de memoria seguro cuestan $8.000 dólares

@Xixauxas @S1XD6XX Jajajaja

@S1XD6XX Lástima que es con cuesta boosteada +30 -10, siendo OTP o máximo 3 champs diferentes y encima empezando desde esmeralda 2 No tiene mérito :)

Grandmaster Corea está bien, pero no es suficiente. Como bien dijo Elm, a mi mitad de viaje ya lo he más que rentado económicamente, pero no en ambición. Podría volver con un millón y seguiría siendo un fracaso de viaje. Cuando yo empecé a jugar a este juego no me importaba el dinero, no me importaba ganarme la vida con el. Lo que me importaba y sigue importándome es calmar la ambición. La única manera de sentirme orgulloso sería que la persona que empezó jugando en un portátil a 15 FPS en Sevilla llegue al rango más alto posible jugando contra sus ídolos en Corea. 280 LPS para Challenger, no voy a defraudar.

@TylerFCloutier TPC-C is pretty outdated

The TPC-C benchmark is a bit silly tbh. I mean sure you definitely want to scale the memory usage, but capping the throughput is silly.

@idoccor @lemire That's because AVX-512 is separated into multiple categories, so saying AVX-512 support only implies 512 bit operations capability (not even register size, AMD's first AVX-512 was using dual 256 but registers). Thank Intel for the fragmentation.

@lemire v4 is really patchy, I found out recently my 7950X supports "AVX512" but not avx512fp16.

On x64 systems (Intel/AMD), we must support a wide range of CPU architectures. Unfortunately, this means software is typically written and compiled for old processors. It is much simpler with ARM, where 64-bit ARM processors are more or less equivalent (with little practical difference in instruction sets). Level 1 (v1). Most Linux distributions are conservative (level v1) and run on almost any 64-bit Intel/AMD processor. You might be surprised to learn that they are more conservative than Microsoft. Level 2 (v2). A few, like Red Hat, Debian 14 (Forky), openSUSE Leap, and Rocky Linux, require v2 (Intel Nehalem 2008 or AMD Bulldozer 2011). This level roughly matches 64-bit ARM (introduced in 2011) and is similar to Windows 11. Level 3 (v3). CachyOS is more aggressive, requiring v3 (Intel Haswell 2013 or AMD Excavator 2015). It mandates AVX2 and 256-bit registers. The v3 level runs on virtually all x64 CPUs built in the last ten years. Your PlayStation 5 is at this level. Level 4 (v4). The real fun begins with v4, which adds full AVX-512 support, the most powerful CPU instructions we have. Only a few distributions fully support running at v4 (CachyOS, Gentoo with custom compilation). To be fair, many distributions offer partial support for higher levels (v3/v4) via glibc-hwcaps, though it is not a complete solution. Why such complexity? 1. Regarding v4, it is mostly Intel’s fault. They invented something powerful but failed to deliver widespread chips. This should improve with upcoming consoles, including Steam Deck successors, which will support v4 by default and may encourage developers (especially game developers) to use it. 2. There is a co-evolution problem. If AMD adds new instructions tomorrow, existing software won’t use them. Upgrading the CPU then brings little benefit. Smooth progress requires lockstep advancement between hardware and software, which has not happened well on x64. What should we do? Linux should follow Microsoft and Red Hat by retiring support for very old x64 processors. These can still be supported by niche or legacy OSes with paid support. The strict minimum should be v2. The primary reason is NOT performance. The primary reason is to allow us to move forward. Make the x64 target more uniform. Make it easy to deploy fancy optimizations. Currently, x64 is a worse platform than ARM in this respect. This is not good. We cannot yet require v4 because of Intel. But once new game consoles ship with v4, it may finally push Intel forward. Game developers should be ready to have fun with the new consoles. Many have no idea what becomes possible with AVX-512 !!!

@lemire Windows is all about backwards compatibility (even if I run some old games better on Linux). I don't see them doing this.

@danNH2006 @Cloudflare Serás tú

@Cloudflare Fraud Alonso is how we call this fraud guy in Spain

AI agents are now ranking among the top hacking teams in the world. Hacker and Cloudflare VP Chema Alonso explains why — and what comes next. Full episode + subscribe → ThisWeekinNET.com

@HotAisle @DellTech Damn, my 42" already feels massive. Tiling ftw, by the way :)

Meet Clint, co-founder of Hot Aisle and the guy who built most of the backend our customers rely on today. He’s now leading what comes next: OpenSushi. Recently, @DellTech moved us into their Enterprise Account Team. The support was already great and this takes it to another level. We now get the same treatment as the big cloud players. I counted 16 people on our org chart. As a nice surprise, I asked if we could send Clint one of Dell’s new 52” monitors. It was just delivered yesterday… and yeah, it’s ridiculously bonkers. They graciously added a new keyboard to pair with it. Mechanics should only work with the best tools. Big thanks to @MichaelDell and the entire Dell team, we couldn’t do this without you. Always professional from day one.

@BriiHD Es lo más probable

Verás tu como echemos a geng para comernos otro 3-0 de la lpl

@magewithnomagic Obviamente no es en todos. De todas maneras, te voy a pasar el EWC, pero no hay clasificatorio, va quien quieren ellos.

@_LTJorge En internacional lleva 2 trofeos seguidos ganados

Chovy vs ganar una serie en una competición sin loser bracket

@magewithnomagic @el30degsw Nadie dice que es malo, si no que chokea en internacional. Y está claro que es así.

@el30degsw Si a vosotros os parece que esto es ser malo o sobrevalorado es para llorar eh

@D0nMaurici0 @Champi_14 Es Kingpin IRL

@Champi_14 El puto aura de Wilson Fisk que tiene don Manute

La G de GENG la pusieron por Guasones. El nombre del equipo coreano viene de Generación Guasona puesto que el CEO era fan del equipo español. De hecho para este First Stand han decidido poner este roster como titular en lugar de los 5 habituales como homenaje.

@Th3AntonioGG F

Oficialmente voy a cambiar de mentalidad, ya no soy Th3Canyon, me convierto en Th3Skewmond Qué barbaridad lo de G2, vaya serie está haciendo... (espero no mufar)

@metatronECP @wallstwolverine ¿Dónde estaba EEUU en la URSS? ¿Y en China, Rumanía, Albania, Yugoslavia, Burkina Faso?

@wallstwolverine 🔴 Hay algo cierto y es que 🇺🇸 nunca ha permitido que el Socialismo Prospere. Si no tiene miedo a qué prospere porque ha perseguido el comunismo? Por qué ha hecho bloqueos continuos a estos modelos? Todavía no sabemos realmente si podría prosperar sin las zancadillas de 🇺🇸

Pablo Iglesias desde Cuba: “Si el socialismo es un fracaso, ¿por qué lo bloquean y no le dejan que fracase solo?”

@feli48183639 @fercardenete Pero si el Nautilus fue de 10 xdd

@fercardenete Lo bueno que se ganó con la aurora de caps y nautilus de labrov jajaja.

No me quiero venir arriba pero es posible que G2 Esports sea el mejor equipo del mundo.

@LethalChicken_ @notch Do it

@notch They're not gonna pull me back in, but they might delay my inevitable switch to linux a little longer.

Wow, people are actually falling for Microslop's promises to change? You don't hesitate when breaking up with a narcissist. Just keep walking away.