Journalist 🗞️ Historian 📚 Writer ✏️ European 🌍 Austrian 🎶 Styrian 🌳 Jedi ☄️ Eldar ✨ Schwoaza ⚽ Dad 👶
Wien, Österreich
Joined June 2010
- Tweets95.6K
- Following2K
- Followers4.4K
- Likes2K
Benjamin Wolf 🇺🇦 retweeted
450 million currently have the right to settle in Amsterdam.
But don’t.
The full 69.5 million population of the UK plus the full 5.5 million population of Ireland currently have the right to work and settle in Brighton.
But don’t.
Reform are being very silly, again.
Benjamin Wolf 🇺🇦 retweeted
Housing affordability has shifted dramatically worldwide. In 1985 a home in the US took 3.7 years of full-time pay to buy. Now you pay 6.4x. Meanwhile, Hong Kong remains the least affordable at over 20 years. Source: buff.ly/Aerd2IV
Benjamin Wolf 🇺🇦 retweeted
🇪🇺🇵🇱 Polska w europejskiej czołówce zadowolenia z życia!
Ranking zadowolenia z życia w UE (skala 0–10). W nawiasach zmiana względem 2021 roku:
🇷🇴 Rumunia - 7,8 (+0,1)
🇫🇮 Finlandia - 7,8 (−0,1)
🇩🇰 Dania - 7,7 (+0,4)
🇦🇹 Austria - 7,7 (−0,3)
🇵🇱 Polska* - 7,7 (+0,2)
🇧🇪 Belgia - 7,6 (+0,1)
🇮🇪 Irlandia - 7,6 (+0,3)
🇲🇹 Malta - 7,6 (+0,5)
🇳🇱 Holandia - 7,6 (0,0)
🇸🇮 Słowenia - 7,6 (+0,1)
🇨🇿 Czechy - 7,4 (+0,1)
🇭🇷 Chorwacja - 7,4 (+0,6)
🇨🇾 Cypr - 7,4 (+0,6)
🇱🇺 Luksemburg - 7,4 (0,0)
🇸🇰 Słowacja - 7,4 (+0,3)
🇸🇪 Szwecja - 7,4 (−0,1)
🇪🇪 Estonia - 7,3 (+0,1)
🇱🇹 Litwa - 7,3 (+0,3)
🇵🇹 Portugalia* - 7,3 (+0,3)
🇪🇸 Hiszpania - 7,2 (0,0)
🇪🇺 Średnia UE - 7,2 (0,0)
🇫🇷 Francja* - 7,1 (+0,3)
🇮🇹 Włochy - 7,1 (−0,1)
🇭🇺 Węgry* - 7,1 (+0,6)
🇩🇪 Niemcy - 7,0 (−0,2)
🇬🇷 Grecja - 6,9 (+0,1)
🇱🇻 Łotwa - 6,9 (+0,2)
🇧🇬 Bułgaria - 6,3 (+0,6)
Według danych Eurostatu za 2025 rok Polska znalazła się w grupie pięciu państw UE z najwyższą średnią oceną zadowolenia z życia.
*Eurostat oznacza dane za 2025 r. dla Polski, Francji, Węgier i Portugalii jako dane o niskiej wiarygodności statystycznej. Dane Polski jednak nie odbiegają wynikiem od ostatnich lat.
Benjamin Wolf 🇺🇦 retweeted
Der Anteil erneuerbarer Energien an der Stromerzeugung steigt immer weiter und wird 2035 bei über 60% liegen.
Benjamin Wolf 🇺🇦 retweeted
'Today, after nearly five years of war, and 22 months of the second Trump administration, the mental map guiding Russian strategy has changed. First, Russia now sees Europe, instead of the United States, as its principal geopolitical adversary, a shift that has driven Russia’s defense establishment to call for an approach known as “castling” to deal with the main threats. Related to this, Moscow now attributes new strategic agency to Europe itself because of the dramatic military modernization taking place in many European countries and the growing interest among some European leaders in developing or extending their own nuclear deterrent. A third shift concerns Russia’s theory of deterrence: Moscow’s traditional understanding of nuclear coercion is giving way to new thinking about nonnuclear strategic deterrence using advanced conventional weapons systems aimed at Europe.
...
One of the most notable trends in Russian strategic thought over the past year has been the emergence of the idea of “castling.” A term borrowed from chess, castling in geopolitics describes the repositioning of forces and resources to optimize a state’s strategic posture. Moscow’s defense experts have used the concept to describe a strategy for dealing with a new reality: for the first time since World War II, they have argued, Europe, rather than the United States, has become Russia’s principal geopolitical adversary. According to this view, European countries are now the principal source of Ukraine’s military capacity, political resolve, and strategic resilience; it is Europe and not the United States that has made Ukraine its proxy and is encouraging Kyiv to pursue Moscow’s strategic defeat.
As a consequence, Moscow must direct coercive pressure at Europe—its elites, its decision-making centers, and its critical infrastructure. By doing so, Russia can instill into Europe’s calculus what theorists in Moscow call a “deterring fear”: a European reluctance to antagonize Russia through further support to Ukraine. Such action, they argue, can also rebuild Russian credible deterrence for long-term competition with the West, which is now centered on Europe.
....
Drawing on this experience, Russian military theoreticians center NNSD on using massive, deep, precision strikes against the adversary’s critical military, political, and economic infrastructure. The strategy, they suggest, should be aimed not only at Ukraine but also at its European supporters. Although the exact contours of NNSD are still emerging, the concept appears to be more escalatory than the hybrid warfare and deniable asymmetric campaigns that Russia is currently pursuing in Europe.'
Benjamin Wolf 🇺🇦 retweeted
Chinese vehicle production is now DOWN about a million cars from its peak late last year ...
Which is remarkable given the scale of China's export boom
1/many
Benjamin Wolf 🇺🇦 retweeted
Great note by @FTAlphaville on the different ways to slice and dice French debt metrics. Firstly, converting it into constant currency to make intl rates more comparable, and then looking at spreads over the overnight funding rates, showing how small the current crisis is compared to Greece in the early 2010s.
This is a seismic flicker at the moment - a warning - but not an earthquake (yet).
In the end, perhaps what matters most is how much of a country’s GDP goes towards interest payments. France has quite a long average maturity (8.5yrs) sparing it a degree of urgency. And despite the last few weeks, France’s EU-leading govt spending and debt/GDP levels, the US and Italy are still set to spend a higher proportion of their economic product on paying off interest.
ft.com/content/9aa4549e-26f4… How nervous are bond markets about France?
Benjamin Wolf 🇺🇦 retweeted
This is a truism about the European bond markets that is written about in every serious note on the topic.
Adam Tooze makes a useful point about bond markets that often gets inverted.
When Germany barely issued new debt (debt brake + surpluses), Bunds became scarce. Bond traders and portfolio managers still need high-quality paper to buy, sell, and hold. So French OATs stepped in as the closest substitute. Demand for French bonds rose, prices went up, yields fell — and France temporarily joined the “high-quality” club alongside Germany.
Scarcity of the safest asset doesn’t mean managers love thin markets. It means capital gets pushed into the next-best alternative until that scarcity eases. Once German issuance picked up and perceptions shifted, France’s spreads widened again.
Benjamin Wolf 🇺🇦 retweeted
Some more AI takes, from several recent conversations:
1. Currently GLM and DeepSeek are much more pleasant to read than any American model.
2. I think there will still be the same order of magnitude job for software engineers in five years.
3. Same for jobs that are seen as box checkers, whether it's coding or accounting or law, are actually human interfaces for dealing with other humans and therefore these jobs do not disappear.
4. AI will continue to get better at learning from patterns that are put in front of it, but this is, as Zenos paradox goes, not a battle that will be fully won. Last miles are hard !!
5. The reason every large company is having difficulty rolling out AI is that they are still believers in the need for deterministic software structures, and until they let go of that and be okay with some mistakes, it will not be fully solved.
6. I suspect fully automated zero employee companies will not happen at scale because there is no marginal benefit and it will get competed away.
7. The number of tokens used will increase by several orders of magnitude beyond what it is today.
8. However, the dollars per token will also drop precipitously from where we are today. In fact, the default assumption for token prices getting cut in half is true, and barring physical or financial system limits, I don't see what will stop this fall. This predicts revenue flattening (still 3 digit growth ofc for a bit).
9. Writing well will continue to be extremely difficult for AI because fundamentally it's an efficient markets problem. Where everybody can write at a particular standard, by definition it is only if you can stand well above it that you command a premium, and this is not just a function of word usage but the entire structural and construction. It's hard for the same reason human world is hard.
10. This is also a world where we will be able to have conversations with every lamppost. A panpsychist world where tokens are consumed and emitted wherever it could plausibly come from, as ubiquitous as electricity.
11. The world is prob going to be one where you can build a model at runtime, fast and bespoke, as needed. The amount of focus on training will therefore increase, not just inference, they're not as cleanly separable.
Some unstructured takes that's taking up mindspace:
1. The frontier labs will continue to make profits and expand their inference empire but the level of profits from frontier models will not be primarily API driven. They'll have to start new businesses and monetise it indirectly.
2. Judgement and taste will turn out to be much harder than anyone thinks. Even as we solve it for specific types of problem, we will not solve it generally that holds temporally (over time) or across domains. Undirected intelligence will exist but not
3. We will not see a jobsocalypse. We will however see jobs change drastically as AI is able to learn parts of jobs very fast and do them. This will seem normal, as per the Douglas Adams rule.
4. The next phase of environments and models will look less like graverobbing company data but working with live data in real settings. More jobs will look like playing a video game for that reason. Everyone's a trader.
5. Regulation is inevitbale. There's too much scrutiny.
6. Mostly we will have to learn to work around AI's flaws (it makes mistakes, it isn't deterministic) rahter than make the flaws work around us through complex software. This is the societal bitter lesson.
7. The challenge for our kids is to live with easy creation while knowing what to create, like for us it was to live with easy consumption but know what to consume (kids have that issue too). It's our job to teach them obsession.
8. I believe the utopian predictions of billions of agents will be true, though also the prosaic predictions that if you see a company in the year 2040, it will look extremely different from companies from the 1960s, but it will also look extremely similar to the companies from 1960, also holds true. A very different reality can look startlingly similar.
9. I know it's fashionable to think that humanities are making a comeback, but I suspect it'll make a comeback as process - how to think better - as opposed to a piece of knowledge where you just know certain things.
10. Software is not dead for the same reason most people don't want to build their own products. Transaction costs will exist. Coase will remain relevant. Hayek will remain relevant. Economics will remain true, yes even for agent swarms.
11. I don't think You Can See Everything is a movie I would like. I base this on the subject matter and having seen Fielder movies before. I do think it'll be a masterpiece. I don't think AI will be able to make one anytime soon. Note I have no problem with AIs making movies per se, I enjoy Marvel.
12. I do think household robots will come but they'll remain shockingly weird and a bit janky and then shockingly commonplace without a major societal upheaval, or at least no more than we had for mobile phones.
13. The GDP growth charts for the world will look normal when zoomed out.
Benjamin Wolf 🇺🇦 retweeted
Fact is, delivering just 50 Taurus cruise missiles and using them on the right targets at the right time would probably save hundreds of IRIS-T/Patriot interceptors.
Hey! What does @bundeskanzler always say? Ukraine doesn't need Taurus KEPD-350 cruise missiles anymore. Sure thing. That's why newly delivered Storm Shadow cruise missiles are literally being used in real time, while they're also rushing to establish a production line in Ukraine.
Benjamin Wolf 🇺🇦 retweeted
Seriously, this is such an insane release, I did not expect this:
Kimi K3.1 Pro beats every listed US frontier model on 6 of 14 benchmarks in Moonshot’s evaluations.
At $2 per million input tokens and $10 per million output tokens, its standard token rates are 50% below Opus 5.5 ($4/$20 !!) and 80% below GPT-6 Astra ($10/$50!!)
In other words: SOTA much cheaper, local running (with enough compute)
Holy shit: Kimi k3.1 pro is on par or even outperforms Opus 5.5 and GPT-6 astra on key benchmarks!!
They just announced K3.1 Pro, a 2.8T-parameter multimodal model with a 1M-token context window that it claims rivals leading proprietary models in coding and agent tasks, with open weights promised by November 4.
China is back!!
Benjamin Wolf 🇺🇦 retweeted
Kimi K3.1 Pro launches at 4th on the Artificial Analysis Intelligence Index scoring 55, making large gains in agentic performance; it will be the leading open-weights model once its weights are published
@Kimi_Moonshot's new reasoning model gains 11 points over Kimi K3 (max, 44). It trails only Claude Opus 5.5 (max with fallback, 58), Claude Sonnet 5.5 (max with fallback, 56), and Muse Wonder Full (max, 56) on the Intelligence Index.
Once the weights for K3.1 Pro are released it will be the highest-scoring open weights model and the gap between the leading open weights and proprietary models on the Intelligence Index will narrow from 12 points to 3.
Key results:
➤ Large gains in agentic performance: Kimi K3.1 Pro scores 3rd on GDPval-AA with 1,846 Elo, up 312 Elo from Kimi K3, behind only Claude Opus 5.5 (1,866 Elo) and Muse Wonder Full (1,857 Elo). Kimi K3.1 Pro also achieves the highest score of any model on Terminal-Bench 4.0. K3.1 Pro scores 67%, up from 13% for Kimi K3 and ahead of Claude Sonnet 5.5 (64%). AutomationBench-AA, which tests agentic SaaS workflows, rises from 58% to 73%.
➤ Gains in other evals are more moderate: Humanity's Last Exam rises from 47% to 52%, CritPt from 23% to 25%, and SciCode from 59% to 61%. AA-Omniscience Index rises from 20 to 30, as the hallucination rate falls from 53% to 44% and accuracy rises from 48% to 51%.
➤ Cost per Task roughly equal to Kimi K3 at $1.98 per Task: K3.1 Pro uses 55K output tokens per Intelligence Index task, 14% more than Kimi K3 at 48K, while its first-party token prices are 33% lower.
Additional model details:
➤ Context window: 1M tokens.
➤ Size: 2.8T total, 104B active parameters
➤ Input modalities: Text and image
➤ Weights and license: Weights currently not published, Moonshot is expected to release the weights soon after launch
➤ Pricing: $2/$10/$0.20 per 1M input/output/cache hit tokens, 33% lower than Kimi K3 on all three rates.
Congratulations to the team at @Kimi_Moonshot on the release!
Benjamin Wolf 🇺🇦 retweeted
Based on this, Kimi K3.1 Pro is ~5 weeks behind the US frontier. That's the shortest the open / closed AI model gap has ever been.
Kimi K3.1 Pro launches at 4th on the Artificial Analysis Intelligence Index scoring 55, making large gains in agentic performance; it will be the leading open-weights model once its weights are published
@Kimi_Moonshot's new reasoning model gains 11 points over Kimi K3 (max, 44). It trails only Claude Opus 5.5 (max with fallback, 58), Claude Sonnet 5.5 (max with fallback, 56), and Muse Wonder Full (max, 56) on the Intelligence Index.
Once the weights for K3.1 Pro are released it will be the highest-scoring open weights model and the gap between the leading open weights and proprietary models on the Intelligence Index will narrow from 12 points to 3.
Key results:
➤ Large gains in agentic performance: Kimi K3.1 Pro scores 3rd on GDPval-AA with 1,846 Elo, up 312 Elo from Kimi K3, behind only Claude Opus 5.5 (1,866 Elo) and Muse Wonder Full (1,857 Elo). Kimi K3.1 Pro also achieves the highest score of any model on Terminal-Bench 4.0. K3.1 Pro scores 67%, up from 13% for Kimi K3 and ahead of Claude Sonnet 5.5 (64%). AutomationBench-AA, which tests agentic SaaS workflows, rises from 58% to 73%.
➤ Gains in other evals are more moderate: Humanity's Last Exam rises from 47% to 52%, CritPt from 23% to 25%, and SciCode from 59% to 61%. AA-Omniscience Index rises from 20 to 30, as the hallucination rate falls from 53% to 44% and accuracy rises from 48% to 51%.
➤ Cost per Task roughly equal to Kimi K3 at $1.98 per Task: K3.1 Pro uses 55K output tokens per Intelligence Index task, 14% more than Kimi K3 at 48K, while its first-party token prices are 33% lower.
Additional model details:
➤ Context window: 1M tokens.
➤ Size: 2.8T total, 104B active parameters
➤ Input modalities: Text and image
➤ Weights and license: Weights currently not published, Moonshot is expected to release the weights soon after launch
➤ Pricing: $2/$10/$0.20 per 1M input/output/cache hit tokens, 33% lower than Kimi K3 on all three rates.
Congratulations to the team at @Kimi_Moonshot on the release!
Benjamin Wolf 🇺🇦 retweeted
⚡️Starshield operations restored to speeds of more than 400 km/h, Ukraine's Defense Ministry says.
"We record a speed of more than 400 kilometers per hour, the connection is stable," the ministry said.
kyivindependent.com/starshie…
Benjamin Wolf 🇺🇦 retweeted
The Beijing Swing mercifully blunted the effects of the Hormuz supply loss through the summer.
And now the unwind may end up blunting the relief of the Middle East's production recovery.
Benjamin Wolf 🇺🇦 retweeted
The Council has agreed its position on EU Inc., a new company law regime to help startups & scaleups do business across the EU.
Companies will be able to register online in less than 48 hours for a maximum of €100 and operate across the single market.
🔗link.europa.eu/YvgkXv
ALT Stylised illustration of office workers using laptops and tablets on coloured square platforms. Text says “EU Inc. How the EU will make it easier for businesses to start, operate and grow across the EU.” Council of the EU logo is featured in bottom left corner.
Benjamin Wolf 🇺🇦 retweeted
The five fastest European companies to hit $100m of ARR are:
5. @ElevenLabs: 20 months
4. @WeAreLegora: 18 months
3. @bfl_ai: 12 months
2. @Lovable: 8 months
1. @microagi: 6 months
The top 4 are also four of the top ten fastest growing companies OF ALL TIME.
Nice
Benjamin Wolf 🇺🇦 retweeted
🇪🇺 UE chce usunąć jedną z największych barier rozwoju europejskich firm.
23 października Rada UE przyjęła stanowisko w sprawie EU Inc., tj. nowej, wspólnej formy prawnej przedsiębiorstw, dostępnej we wszystkich 27 państwach Unii.
Projekt przewiduje rejestrację firmy online w maksymalnie 48 godzin, za nie więcej niż 100 euro i bez wymaganego kapitału minimalnego. Ma też ułatwić pozyskiwanie inwestorów i ekspansję zagraniczną.
W UE obecnie funkcjonuje ponad 60 krajowych form prawnych spółek. Rozdrobnienie przepisów zwiększa koszty i utrudnia europejskim przedsiębiorstwom osiąganie efektu skali, z którego korzystają firmy na ogromnym rynku amerykańskim.
EU Inc. to część szerszej strategii „One Europe, One Market”, której celem jest usuwanie barier wewnętrznych i pogłębienie integracji gospodarczej UE do końca 2027 roku.
