Flauneu

@vshih2 His comments related to Taiwan & geopolitics. Not science. I don't know what you're talking about.

I have always thought part of the right is very compatible with CCP interests. One side would like to denigrate science and knowledge domestically while the other side wants to dominate sciences globally and export tech products to those incapable of producing them

@DesmondShum @APHClarkson Iran is not a threat to Europe. And this war was started for Israel and not for Europe. You're pretty low IQ, desmond.

The United States would be unable to project power on a global scale without access to bases and supply chains in the EU, UK and Japan

@iggnacy @Olowek25 Stupid analysis. The main leverage Iran has is the skyrocketing cost of oil. This makes it different from Iraq war. Also, the US does not have the capability to open the Strait of Hormuz. Mickey Mouse analysts in Poland with a US worship. Bearish sign.

Czy Iran może długo blokować Cieśninę Ormuz?? Czy USA mogą ją odblokować?? Rozmawiałem z dr. Łukaszem Przybyłą @Olowek25 z Akademii Sztuki Wojennej, autorem książek też o wojnach na Bliskim Wschodzie. Kluczowe wnioski z rozmowy: => Iran stracił już znaczną część potencjału balistycznego – te pociski są duże, łatwe do wykrycia i zniszczenia. Ale drony to zupełnie inna historia. Mogą nękać region długo => Amerykanie mają zdolność do ochrony ruchu w Cieśninie. Skoro zdegradowali irańską obronę przeciwlotniczą, mogą utrzymywać bezzałogowce nad południowym Iranem 24/7. Systemy big data i AI (np. Palantir) tworzą wzorce pozwalające lokalizować magazyny, rampy startowe itd => Czy USA są zaskoczone przebiegiem wojny? Wątpliwe. To dowództwo przez 20 lat prowadziło wojny w Iraku i Afganistanie. Długo szykowało się to tej wojny. Ale publiczność chce efektów „na kciuk", lajki, szybko, raz dwa => To jest ostatecznie pojedynek na wytrzymałość: polityczną po stronie USA i reżimową po stronie Iranu. Szyici mają cierpienie wpisane w fundamenty wiary. Ale Iran nie sprzedaje ropy, nie sprzedaje gazu, jest pod ciągłą presją bombardowań Cała rozmowa, więcej wniosków 👉👉👉pb.pl/swiat-oczekuje…

If you happen to be rate limited again, just use #Claude for free through Amazon customer support

@Geenimetsuri @StefanFSchubert In practice, the renting market is even more segregated than the owner market in Stockholm. To get a rent-controlled apartment in the core areas, you need to wait >25 years. People who can afford to wait that long have good finances already. It's not poor people who live there.

@StefanFSchubert Err...didn't you just prove the opposite?

How market rents would change rents in Stockholm. Rent would double in desirable parts of the city centre, but wouldn’t change at all in less desirable suburbs. Rent control is a very ineffective way to reduce inequality.

@StefanFSchubert Liberals are smarter in every society and the US is no different.

Pride in being American and support for Israel used to be consensus views in the US, but in the last few decades, they’ve quickly become highly polarising. While Democrats' views have shifted quickly, Republicans' haven’t changed. update.news/p/when-the-env…

@Jsevillamol Is Merkel the worst German chancellor in modern history?

I am so mad Germany decommissioned 20GW of nuclear power since Fukushima. This will go down as one of the greatest climate change and AI fumbles of history.

Announcing DreamDojo: our open-source, interactive world model that takes robot motor controls and generates the future in pixels. No engine, no meshes, no hand-authored dynamics. It's Simulation 2.0. Time for robotics to take the bitter lesson pill. Real-world robot learning is bottlenecked by time, wear, safety, and resets. If we want Physical AI to move at pretraining speed, we need a simulator that adapts to pretraining scale with as little human engineering as possible. Our key insights: (1) human egocentric videos are a scalable source of first-person physics; (2) latent actions make them "robot-readable" across different hardware; (3) real-time inference unlocks live teleop, policy eval, and test-time planning *inside* a dream. We pre-train on 44K hours of human videos: cheap, abundant, and collected with zero robot-in-the-loop. Humans have already explored the combinatorics: we grasp, pour, fold, assemble, fail, retry—across cluttered scenes, shifting viewpoints, changing light, and hour-long task chains—at a scale no robot fleet could match. The missing piece: these videos have no action labels. So we introduce latent actions: a unified representation inferred directly from videos that captures "what changed between world states" without knowing the underlying hardware. This lets us train on any first-person video as if it came with motor commands attached. As a result, DreamDojo generalizes zero-shot to objects and environments never seen in any robot training set, because humans saw them first. Next, we post-train onto each robot to fit its specific hardware. Think of it as separating "how the world looks and behaves" from "how this particular robot actuates." The base model follows the general physical rules, then "snaps onto" the robot's unique mechanics. It's kind of like loading a new character and scene assets into Unreal Engine, but done through gradient descent and generalizes far beyond the post-training dataset. A world simulator is only useful if it runs fast enough to close the loop. We train a real-time version of DreamDojo that runs at 10 FPS, stable for over a minute of continuous rollout. This unlocks exciting possibilities: - Live teleoperation *inside* a dream. Connect a VR controller, stream actions into DreamDojo, and teleop a virtual robot in real time. We demo this on Unitree G1 with a PICO headset and one RTX 5090. - Policy evaluation. You can benchmark a policy checkpoint in DreamDojo instead of the real world. The simulated success rates strongly correlate with real-world results - accurate enough to rank checkpoints without burning a single motor. - Model-based planning. Sample multiple action proposals → simulate them all in parallel → pick the best future. Gains +17% real-world success out of the box on a fruit packing task. We open-source everything!! Weights, code, post-training dataset, eval set, and whitepaper with tons of details to reproduce. DreamDojo is based on NVIDIA Cosmos, which is open-weight too. 2026 is the year of World Models for physical AI. We want you to build with us. Happy scaling! Links in thread:

@AcerFur Ignore the haters. I found out about Loreen's latest tune thanks to you and it's a banger. You're one of the few AI-related accounts that actually tweet interesting stuff about non-tech topics.

@mbolotnikova YIMBYs want massive density, praising the skyscraper hell of Manhattan. Anyone interested in beauty will come to a realisation that NIMBYism, to a certain extent, is required. Just think about the concept of "human-scale architecture". NIMBYism is simply unavoidable for that.

Why is the modern American built environment overwhelmingly ugly & soul-deadening? And should YIMBYs care about it? Our housing crisis is a nightmare for millions of Americans, but could it also, maybe, be a rare invitation to rebuild the way we live? threading gift link!

@MarekJozefiak A captive domestic market is easy. The real test is exports. So far there's almost nothing there. Will the Turkish taxpayer subsidise a carmaker without international success? Maybe - look at how long Malaysia allowed Proton to go on. But that doesn't mean it's a good model.

"Turecka Izera" tym różni się od polskiej, że powstała i ma się dobrze. Państwowy Togg w 2025 roku był numerem jeden na chłonnym tureckim rynku elektryków, a udział e-samochodów wyniósł tam prawie 18% - dwa razy więcej niż w zamożniejszej Polsce. dailysabah.com/business/autom….

This chart documents one of humanity’s greatest achievements, in my view. We just lived through the fastest population growth in history. It would have been impressive if food supplies had merely kept pace — but on every continent, they grew even faster than the population.

@iggnacy You're missing the fundamental issue. There is no coherent European identity. No amount of "innovation" in politics will solve this. Europe was strong when *individual* European countries had a higher share of world demography, and when East Asia was poor. This is structural.

Problem tu dyskutowany - taki, że UE nie potrafi wykazać determinacji w reakcji na nowe wyzwania - polega jak sądzę na tym samym, na czym słynny dylemat innowatora opisany przez Claytona Christensena. Wielkie korporacje nie są zdolne do reakcji na nowe zjawiska. W oryginale te nowe zjawiska to technologie i produkty. Tutaj jest to nowe zagrożenie. Ale problem jest w istocie ten sam. Dzisiejsza Europa jest zorganizowana na zasadzie korporacji. Rządzona przez anonimowe procedury, jak sprawnie działającą maszyna nastawiona na określony cel. Zbudowana do radzenia sobie z problemami XX wieku. Maszyna, w której trudno zidentyfikować odpowiedzialność. Nowe zagrożenie wymaga podjęcia dużego ryzyka. Nowych idei, schematów, poświęcenia. A to wymaga odpowiedzialności. Tego w maszynie korporacyjnej nie ma. W dylemacie innowatora problem rozwiązywany jest tak, że nowym wyzwaniom stawiają czoła nowe podmioty. Polityka też wymaga innowacji. Stąd takie konstrukty jak "koalicja chętnych". Ale docelowo będzie to oczywiście dużo, dużo trudjiejsze niż w biznesie. To nie brak takich czy innych liderów jest tu wyzwaniem.

@peterwildeford Litt says it does. x.com/littmath/statu…

Another question will be how much does progress in software engineering, which METR measures, also connect with other tasks. METR had previously found other tasks doubling at a similar rate, but I think it's time to revisit.

Recent METR data shows ~4.4mo doubling for 50% reliability, with frontier models now completing median tasks up to ~5 hours. Which is crazy! 80% reliability fits similar ~4.5mo, but w/ more variance. Reliability gains may be plateauing somewhat? Worth monitoring.

In 2006, two academics, John Mearsheimer and Stephen Walt, wrote a paper called “The Israel Lobby,” which argued that a loose group of Israel supporters used their connections and financial resources to tilt US foreign policy in Israel’s direction. One of them, Stephen Walt, was academic dean of The Harvard Kennedy School, which put him at the pinnacle of his profession. The savage response to the paper was like nothing academia had ever seen. We can now report that Jeffrey Epstein helped coordinate the counterattack with Alan Dershowitz, while Larry Summers was Harvard president. And Les Wexner was one of the largest donors to the Kennedy School, if not the largest. Epstein controlled Wexner’s money. In other words, a loose group of powerful supporters of Israel used their financial resources and connections behind the scenes to destroy the reputations of two academics who wrote a paper arguing that a loose group of Israel supporters was using their connections and financial resources to shape American foreign policy. Story here dropsitenews.com/p/jeffrey-epst…

@StefanFSchubert @ahistoryinart You should really write up a travel guide for Stockholm. You probably have a more refined taste than most people.

@ahistoryinart Nice, thanks for posting. He was also an art collector. His former home is a beautiful museum, my favourite in Stockholm. waldemarsudde.se/en/

It's only recently that I've come across work by Prince Eugen of Sweden, the wonderfully named Duke of Närke, and the youngest son of King Oscar II and Queen Sophia. He trained under Leon Bonnat and Puvis de Chavannes - this is a scene at Lövsprickning, Balingsta in 1887.

@tugot17 @Jukanlosreve You need to read up on HBD: zixuanma.blog/p/22-million-g…

@Jukanlosreve how come no other country has ANY industrial policy to at least try to counter this 🥲

Goldman Sachs: “China has already dominated the robot supply chain: gearing up for mass production in the second half of 2026 — securing annual production capacity of 100,000–1,000,000 units” • According to a recent survey by Goldman Sachs of major Chinese robot component makers such as Sanhua (三花), the industry views the second half of 2026 as the time when large-scale mass production of “humanoid robots” will begin in earnest. This is presented as a key benchmark for gauging investment timing across the robot industry value chain. 1. A clear timeline: It expects the inflection point for large-scale mass production across the supply chain to be the second half of 2026, which will serve as an important temporal anchor point for investment in the robot industry. 2. Coexistence of confidence and risk: Component and module suppliers are adopting an “invest first, demand later” strategy through real-economy investments such as securing land, expanding factories, and building production lines. This both reflects strong confidence in a surge of orders from key OEMs such as Tesla and entails the risk of overcapacity in production facilities if demand falls short. 3. Reconfiguration of the value chain: Suppliers are shifting from simple component sales to a focus on “module/system integration.” In particular, companies that provide integrated actuator/sensor solutions and possess global production capacity are expected to move up the value chain and gain bargaining power. 4. Points to watch: Goldman Sachs presents two key events over the next 1–2 years to confirm the industry’s turning point. (1) Tesla Optimus, 3rd generation: scheduled to be unveiled in February–March 2026, (2) The 2026 shipment/order targets of global robot companies: expected to be announced in late 2025 to early 2026. These two will be the critical litmus test for whether the optimistic outlook translates into actual orders. In summary, Goldman Sachs’ report describes China’s humanoid robot supply chain as “everything is ready except for the very last step.”

@arctotherium42 Chinese people know far more about the US than vice versa. Also, many of the guy's examples of supposedly "misguided Chinese beliefs" are actually quite rational. The trade war is seen as an attempt to strangle - or slow - China's economy, even in countries outside of China.

It is a bad thing that the two most important countries in the world understand each other so poorly. This is easy to see from the US side (we tend to rely on diasporoids with axes to grind), not surprising it's true of China too.

"Humans aren’t very efficient movers—until you put us on a bicycle, when we become some of the most energy-efficient land travelers in the animal kingdom."

@matigpt @daniel_mac8 METR publishes a different methodology for 80% success rate and the top model does 26 minutes as of writing. So if you have high standards, the SOTA models are still very far away from hours of high-quality work.

@daniel_mac8 I have yet to see any proof. My tests tell a whole different story. It is tough to push ANY model out there beyond 1 hours. 30 hours? If that was true there should be people out there reporting 10, 15, 20 hour runs. But there are none.

Doubt Anthropic was lying when they said they observed Sonnet 4.5 working for 30 hours. METR requires a 50% success rate for the duration. And Anthropic never said it did the task correctly.