Den säkert underbara fadern, sonen, brodern och EU-parlamentarikern som röstat för att en digitalbox inte får ha opt-out för autoavstänging må få en klapp på kinden - av ett absolut fucking järnrör.
Svenska
Andreas Eriksson
1.2K posts
@Stef14211267 @BSonnert Väldigt mycket hänger på morgondagens match mellan City och Palace ifall det blir oavgjort idag. Målskillnaden är lite för spännande just nu. City kan absolut ta Villa och Bournemouth med ett par bollar om de har vittring och plusmål efter morgondagen.
Svenska
Ungefärliga sannolikheter vid olika resultat.
Om West Ham vinner:
West Ham relegerade 44%
Arsenal vinner ligan 46%
Om kryss:
West Ham relegerade 72%
Arsenal vinner ligan 57,5%
Om Arsenal vinner:
West Ham relegerade 81,5%
Arsenal vinner ligan 86,5%
Förståeligt som neutral att West Ham-vinst är otroligt för spänningen. Själv tar jag gärna tvåan, orkar inte mer spänning 😅.
Svenska
@ivanfioravanti @micheltamanda June? More like October-November at best.
English
@micheltamanda They are keeping RAM for M5 Ultra in June 💪
English
You snooze you lose, pretty soon there will be zero inventory.
Basic Apple Guy@BasicAppleGuy
Apple removed the 256GB memory option for the M3 Ultra Mac Studio. Only configuration is now 96GB.
English
One of the weirdest players we’ve ever had ngl
Don Draper@arsenalbarclays
Not bad for a debut season
English
@philipsemmler @hofmannjesper Ja och nej, Gyökeres använde ju en Saka som ofrivilligt bollplank när han smekte den i krysset tidigare under säsongen 😅
Svenska
@hofmannjesper Första gången någon av de assisterat till den andre på hela säsongen.
Svenska
Vi har inte sett mkt av firma Saka - Gyökeres den här säsongen. Otroligt fint för Arsenal om det lossnar nu.
Svenska
@TheAhmadOsman they'd never put 1TB in the Studio, that's Pro/Ultra territory lol. safe bet, Apple's not cannibalizing their own stack
English
@greatscottdev @rot13maxi @QuixiAI 397b for general purpose, sure. Given the fact Coder-Next still holds its own, 3.6 Coder-Next 120b ~a9 would probably where we can settle for serious coding quality within a 10k budget.
English
@rot13maxi @QuixiAI Not really, 397 is the most useful option.
English
@PEKtoBKK @IsaCNN @carlbildt If by ”sorted”, you mean ”attacking the Trump Regime with absolute determination”, you are absolutely right.
Trump and the US needs to pay the price. Revamping their democracy from the core and up. That’s the only way they can be trusted again by the weststern civilization
English
@IsaCNN @carlbildt That's not technically how things work. The group most effected by the problem actually needs it to be sorted
English
“We didn’t have a problem with the Strait of Hormuz; that was created by the war that the U.S. Trump administration launched. So they have created a problem, and it is their main responsibility to sort out the problem that they created”— tells me @carlbildt
English
@tamira1948 @3YearLetterman @tim_cook @Tim_Walz @TimTebow @UrbansPintHouse @GropingIn @HomeGoods @PublixJobs @Kash_Patel @JackDaniels_US @Whitesnake That is not a dummy of Bill Gates 🤦♂️
English
@3YearLetterman @tim_cook @Tim_Walz @TimTebow @UrbansPintHouse @GropingIn @HomeGoods @PublixJobs @Kash_Patel @JackDaniels_US @Whitesnake It's Bill Gates dummy!
English
🚨 BRAKING 🚨
Apple’s founder and CEO, Tim Cook, is stepping down. This is a dark day for America, as he literally invented the computer, the iPod, and the iPhone
English
@tteokl_ @HarshithLucky3 Yes, Claude is a distillation of basically everything online.
English
@devteamdrew @0xSero Router-weighted Expert Activation Pruning
English
@jt_martin @kimmonismus Qwen3.5 27b is what Gemma 4 31b is up against. Don’t compare it to any MoE.
English
This chart is the whole story of where AI model economics are heading.
A 31B parameter model matching or beating models that are 12-22x its size. Gemma 4 at 1452 Elo sitting right next to Qwen 3.5 at 1450 with 397B parameters, and Kimi k2.5 at 1454 with 1.1 trillion parameters.
The inference cost implications are staggering. Running 31B parameters costs a fraction of what 397B or 1.1T costs in compute, memory, and power. If quality converges at the top, the winner is whoever runs the smallest model at equivalent performance.
This is why the training cost arms race narrative is misleading. Yes, frontier models cost $500M+ to train. But the distillation and architecture improvements that let 31B compete with 397B are where the real value creation happens.
Google's Gemma strategy is Meta's Llama strategy with better execution. Give away the model. Win the ecosystem. Monetize through cloud and search.
English
Gamma 31b model outperforming Qwen 3.5 397B is nuts to me.
Google DeepMind@GoogleDeepMind
Available in four sizes: 🔵 31B Dense & 26B MoE: state-of-the-art performance for advanced local reasoning tasks – like custom coding assistants or analyzing scientific datasets. 🔵 E4B & E2B (Edge): built for mobile with real-time text, vision, and audio processing.
English
@LottoLabs Good to know. Not even loading on my Strix Halo no matter what I do.
English
Lots of people outing themselves by not knowing qwen3.5 27b is arguably better and been out for 4-5 weeks already
Nick O’Neill@chooserich
It cannot be overstated how big this news is. Being able to run a frontier AI model on your own hardware means token costs are effectively free. AI will be as ubiquitous and cheap as the internet itself. This is a dagger in OpenAI and Anthropic.
English
Best models to run on your hardware:
—— 64 GB ——
- Qwen3-coder-next-80B-4bit (coding, Claude code, general agent)
- Qwen3.5-122B-reap: (browser use, multimodal, tool calling, general agent)
—— 96 GB ——
- GLM-4.6V (multimodal and tool calls)
- Hermes-70B (Jailbroken)
- Nemotron-120B-Super: (openclaw)
- Mistral-4-Small (general agent)
—— 192 GB ——
All these are excellent top tier LLMs and approach sonnet in capabilities
- Step-3.5-Flash
- Qwen3.5-397B-REAP
- MiniMax-M2.5 (soon M2.7)
- GLM-4.7-Reap
0xSero@0xSero
Best models to run on your hardware level I'll be doing this every week, I hope you guys enjoy. ---- 8 GB ---- Autocomplete for coding (like Cursor Tab) - huggingface.co/NexVeridian/ze… - huggingface.co/bartowski/zed-… Tool calling, assistant style - huggingface.co/nvidia/NVIDIA-… ---- 16 Gb ---- Here things get better: Multimodal - huggingface.co/Qwen/Qwen3.5-9B - huggingface.co/Tesslate/OmniC… - huggingface.co/unsloth/Qwen3.… ---- 24 GB ---- - The best model you can get (thanks Qwen) huggingface.co/Qwen/Qwen3.5-2… - Great model (strong agents) huggingface.co/nvidia/Nemotro… - Mine hehe huggingface.co/0xSero/Qwen-3.… I'm doing a weekly series
English
@MadeleineLidman Det är ingen uthängning om det är media som gör bort sig. Det är snarare gratis reklam. Det står också regionchef på hennes kläder.
Svenska
Till den stackars anställda på
Max som nu hängs ut i media för att hon råkade servera fel hamburgare: det är sådant som händer, du har folket på din sida.
Till den journalist som ägnar sig åt maktmissbruk - skäms!
Svenska
@MaxVillman Google, Anthropic and OpenIA should probably just stfu.
English
@Coooter @Israel2252 Yes, that is 100% empty. You can see it in the original video.
English
WATCH ALEX PRETTI’S LEFT HAND:
🫱Left hand hand goes down empty.
🫱 🔫 Left hand comes up with a ___.
English
@Minette58 @Erkanperkan4 @SwedishPM Nej… USA är enda landet utom Danmark som har militär bas på Grönland. USA är välkomna att bemanna den. Sannolikt även välkomna att öppna fler militärbaser på ön. USA’s huvudintresse ligger dock ej i försvaret. De vill åt mineralerna, annars hade de redan öppnat fler baser.
Svenska
Samtalade nyss med Danmarks statsminister Mette Frederiksen tillsammans med våra nordiska kollegor. Vi står bakom Danmark och Grönland.
Svenska
@Afterseven @WarMonitor3 This is a Russia-Venezuela-USA situation. I would be quite embarrassing if Denmark won 🤕
English
@WarMonitor3 Well there's only gonna be one winner. and it ain't Denmark and it ain't the EU
English
A potentially deadly confrontation in the Atlantic is upcoming.
English
