maggy

Norwegian pokerplayer

[Remix News & Views @RMXnews]

🇫🇷 A store in France has put security tags on every Algerian and Moroccan World Cup jersey.

On the French ones? Nothing.

"Coincidence? I don't think so."

[maggy @magne87]

@morten_00 @mariusfla Løfter hele opplegget flere hakk!

[Benjamin Bringsås @benjabrin]

"Vedum har et poeng: de sitter altså og forhandler om et statsbudsjett for Norge i 2026, men SV har hovedfokus på Midtøsten og MDG snakker om en tidsplan for avvikling av olje og gass på norsk sokkel(!). Som Vedum sier: man nevner ikke Trøndelag eller Agder én gang, men nevner steder i Midtøsten 50 ganger. Dette er useriøst uansett hvor man står politisk."

Skattet sitat verdt å dele.

[Pepsigro @PepsiGro]

Iblant kommer det et glimt av lys, slik at man kan se hvor folk faktisk står.

I dag ser man det blant en del av de «velintegrerte» andregenerasjonsjonsinnvandrerne. De kulturelle føringene ligger der fortsatt. Jøder er slemme. Muslimer er ofre.

[Tommy Robinson 🇬🇧 @TRobinsonNewEra]

The supposed journalist, Mr FAFO, who was fake killed by Israel multiple times for Pallywood purposes by Hamas, was actually killed at the end of the war as hamas fought groups who wanted hamas out.

This is his memorial in "Palestine".

Joking, it's Oslo, Norway 🤦🏻‍♂️

[Rune Østgård @enur72]

Kommer til en barnehage nær deg 🔥🔥🔥

[Anon Bitcoin Hodler @anon_hodler]

Først lager de problemet, så selger de oss løsningen… i form av mer overvåking, mindre frihet.

[Bernt Berg-Nielsen @berg_bernt]

Ikke ofte jeg tar investeringsråd fra Rødt, men @mimirk sin anbefaling i dagens DN om å løpe langt vekk når powerpoint-med-bærekraftsmål kommer frem er 🎯

[maggy @magne87]

@LiseSorensen_ La meg gjette: noe grønt!

[Pepsigro @PepsiGro]

Mørkt

How life expectancy in the UK fell into a Soviet-style slump

https://t.co/sdwONBSK80

[henrikbeckheim @henrikbeckheim]

Erling Haaland ringer opp det tidl. israelske gisselet Omer Shemtov 😮 Haaland gjør en mye bedre jobb for Norge, for våre internasjonale relasjoner, og ifht. å bare være et medmenneske, enn det både Støre og Barth Eide klarer.

[maggy @magne87]

@Deltidspoker Dessverre flere vinnere her. Media og advokatene de to største

[maggy @magne87]

Beste forklaringen på hva som skjer i USA jeg har lest på norsk!

[Helge Morset @HelgeMorset]

Wtf er dette? En nøktern analyse? I Aftenposten??

https://t.co/lXNWVVhiRe

[David Bernstein @ProfDBernstein]

Why does @piersmorgan simply assume that Hamas-supplied casualty figures are (a) accurate; and (b) reflect civilian deaths? Here's what the Israeli newspaper Ma'ariv reports (translated to English):
The military operation "Oz ve Harb" carried out at night in Gaza constitutes one of the largest counter-terrorism operations in military history. In a coordinated operation that lasted just a few minutes, hundreds of terrorists and senior Hamas figures were eliminated thanks to extraordinary cooperation between the Shin Bet and the Air Force, led by "the two Bars" - Ronen Bar and Tomer Bar.
The official name of the military operation carried out last night in Gaza is "Strength and Sword," but the real name should reflect Operation "Double Bear." The operation last night is one of the largest counterterrorism operations in military history. According to Palestinian reports, over 300 terrorists, including senior Hamas figures, were eliminated within a few minutes. The operation was planned by the Southern Command, the Air Force, the Intelligence Division, and the Shin Bet.
The operation was supposed to provide an opening move similar to "Operation Beepers" in Lebanon in September of last year. Unlike the Beepers operation, in which the bomb was planted in advance in the devices that the terrorists were holding. But in the current operation, the Shin Bet was required to prepare a particularly complex intelligence plan. The basis of the plan was the following: preparing a list of dozens of Hamas terrorists of various ranks and with wide circles of knowledge in the organization. The second thing was to locate each of them geographically at the time of the attack. The Air Force received the NAP of each terrorist who was targeted for elimination, and each NAP was inserted into the bomb intended for him

[Elon Musk @elonmusk]

As predicted

[maggy @magne87]

@RettPaaMedia Mener du Farming Simulator?

[Mikkelickski @mikkelickski]

«…..mente sensor at bruken av «male og female», altså «han- og hunkjønn» var problematisk og at det ga uttrykk for en «biologisk deterministisk» forståelse av kjønn som ikke var forenelig med emnets kompetansemål.»

https://t.co/ltxFh6bwms

[Arnaud Bertrand @RnaudBertrand]

This 👇 is arguably an even bigger Sputnik moment for China than the 6th generation fighter jet: a Chinese AI Model called DeepSeek v3 rivals - and often surpasses - the latest ChatGPT and Claude models in pretty much all respects for a tiny fraction of the training cost (only $5.5m), and it's open sourced (meaning anyone can use, modify, and improve it).

The fact that it's so cheap to train is particularly important as it completely changes the game of who can participate in advanced AI development. Up until now, the assumption was that you needed hundreds of millions or even billions of dollars to train such a model, yet DeepSeek did it with just $5.5m, a sum of money accessible to just about any startup anywhere. Concretely, this means that DeepSeek has just proven that serious AI development is not limited to tech giants.

And their model is not only cheap to train, it's also extremely efficient to run. They use an architecture called Mixture-of-Experts (MoE) where, while their full model has 671 billion parameters (which is huge), it only uses 37 billion at a time. To compare, Meta has 405 billion parameters in their latest Llama3.1 model and uses all 405 billion at a time. DeepSeek V3 is more than 10 times more efficient, yet performs better than Llama3.1 at almost all benchmarks (English, Math, Coding, etc.).

DeepSeek V3's performance at key benchmarks is impressive across the board:
- Crushes advanced math problems (90.2% on MATH-500, vs 78.3% for Claude-3.5-Sonnet and 74.6% for GPT-4o)
- Excels at coding (82.6% on HumanEval, vs 81.7% for Claude-3.5-Sonnet and 80.5% for GPT-4o)
- Can process huge amounts of text at once (128K tokens, roughly equivalent to 100,000 words in English)
- Processes text at 60 tokens per second, about twice faster than GPT-4o

And the craziest part is that it's open-source, meaning that:
- Anyone can download and study the code
- Developers can modify and improve it
- Companies can integrate it into their products without paying API fees
- The entire AI community can learn from it

Lastly, this obviously comes during an interesting context in China-US relations where the US is doing its utmost to prevent China from progressing technologically, especially in AI. As such, this is an absolutely beautiful response by China: "despite all your restrictions, we just built a world-class AI model for 1% of your cost, made it more efficient than anything you have, and open-sourced it for the whole world to use."

It's also a triumph of brains over money and raw power: with its restrictions the US placed China in a situation where it had to use resources more intelligently. As the saying goes "necessity is the mother of inventions"... And here we now are: China may have just changed the rules of the game forever, democratizing the very technology the US tried to restrict and proving, once more, that human ingenuity always finds a way.

[Defiant L’s @DefiantLs]

Jake Tapper shocked to learn Harris didn't outperform Biden.

“Holy smokes. Literally nothing?”

https://t.co/m7vKN06gqO

[Daniel Vassbund @Smoothedan]

Øystein Stray Spetalen er, som vanlig, helt spot on. De utallige milliardene politikerne våre kaster bort på «det grønne skiftet», er et hån mot skattebetalerne. 💸