Marcel Weiß

@marcelweiss.bsky.social

460 Followers 383 Following

Strategy Analyst & Publizist - FAZ, neunetz.com, neunetz.fm, Lieblingsfarbe Blau

Pre-merge, this took 5.9 billion uncached input tokens, 690 million output tokens, and 72 billion cached input token reads — around $165,000 at API pricing. By hand, I think this would've taken 3 engineers with full context on the codebase about a year, during which time we wouldn't be able to improve Node.js compatibility, fix bugs, fix security issues or implement new features. We never would've done that. The realistic alternative was to do nothing and keep fixing the bugs at the top of this post forever.
South Korean government and companies announce massive investments in chipmaking and AI data centers to boost technology infrastructure.
The chart compares performance metrics of AI models Claude Fable 5, Opus 4.8, GPT 5.5, and Gemini 3.1 Pro across various tasks. Fable at 0.
itsa me
Fables Vol. 18: Cubs in Toyland
Moodymann dances on skates to Prince.
— for the people saying things like “I’m a PhD from xyz uni and I’m telling your LLMs are just stochastic tools that make everything up and the world will fall apart if you use them”, I’m here to tell you that you are out of date. The world of software engineering has changed dramatically in the last few months. The world of IT security and maintaining software in the face of the flood of reports has completely and utterly changed just in the last few weeks. Anything you learned about this stuff last year might as well be from another planet. Oh, btw, I also did a PhD in CS, and yes, I actually did quite a lot of work on neural networks, and yes, all my knowledge from that time in my life is utterly out of date too. Also, nobody actually knows if human
The OpenAI achievement was “remarkable” but likely involved substantial compute resources, said Ursula Martin, a mathematician and computer scientist at Oxford University and an author of the declaration, in The New York Times interview. She suggested that similar quantities of equivalent effort from human mathematicians would have probably solved the problems in the same way—and she cautioned that mathematics is also about the “cultivation of ideas, understanding, judgment and human insight” beyond solving problems.
OpenAI uploaded a research paper describing its AI model’s mathematical proof along with commentary from independent mathematicians. But the company did not disclose information about prompts, AI training data, and the amount of computational resources used to solve the mathematics problems in question, said Rodrigo Ochigame, a historian and anthropologist of computing and artificial intelligence at Leiden University and another author of the declaration. “The AI model is proprietary and unavailable to anyone outside the company,” Ochigame told The New York Times. “We get a flashy promotional video, while basic information needed to assess the scientific meaning of the result is kept secret.”
“The tech industry proceeds in accordance with commercial logic, which is antithetical to the values of mathematics,” said Michael Harris, a mathematician at Columbia University and an author of the declaration, in a New York Times interview. He also spoke of the declaration attempting to “recover control of the narrative about the values and goals of mathematics from the AI industry.”
With the advancement of AI capabilities, AI reviewers are beginning to be deployed in scientific peer review, yet their capability and credibility remain in question: many scientists simply view them as probabilistic systems without the expertise to evaluate research, while other researchers are more optimistic about their readiness without concrete evidence. Understanding what AI reviewers do well, where they fall short, and what challenges remain is essential. However, existing evaluations of AI reviewers have focused on whether their verdicts match human verdicts (e.g., score alignment, acceptance prediction), which is insufficient to characterize their capabilities and limits. In this paper, we close this gap through a large-scale expert annotation study, in which 45 domain scientists in Physical, Biological, and Health Sciences spent 469 hours rating 2,960 individual criticisms (each targeting one specific aspect of a paper) from human-written and AI-generated reviews of 82 Nature-family papers on correctness, significance, and sufficiency of evidence. On a composite of all three dimensions, a reviewing agent powered by GPT-5.2 scores above each paper’s top-rated human reviewer (60.0% vs. 48.2%, p = 0.009 ), while all three AI reviewers (including Gemini 3.0 Pro and Claude Opus 4.5) exceed the lowest-rated human across every dimension. AI reviewers’ accurate criticisms are also more often rated significant and well-evidenced, and surface a distinct 26% of issues no huma
Genau dafür wurde Starlink gebaut. Nicht weil Musk zuerst Internet aus dem All anbieten wollte, sondern weil die Falcon 9 ohne einen massiven Eigenkunden ihr Kostenpotential nicht ausspielen konnte. Und so sind wir da gelandet, wo wir heute stehen. Heute betreibt SpaceX 9600 aktive Satelliten im niedrigen Erdorbit. Das sind rund zwei Drittel aller aktiven Nutzlastsatelliten. Die Zahlen sprechen für sich: 10,3 Millionen Abonnenten in 164 Ländern. Mehr als 16 Millionen monatlich aktive Geräte über das Direct-to-Cell-Angebot in 32 Ländern, in denen 1,7 Milliarden Menschen leben. Das geht aus dem SpaceX-Börsenprospekt hervor. Was Europa dagegenhält Die europäische Antwort heißt IRIS². Geplant sind 290 Satelliten, erste Dienste 2029, voller Betrieb 2030 laut (EU Space Programme). Eutelsat-OneWeb, der einzige operative europäische Satelliteninternet-Anbieter, betreibt rund 650 Satelliten in 1200 Kilometer Höhe; das Geschäftsmodell richtet sich an Behörden und Großkunden. Für die Versorgung eines Bauernhofs im Sauerland oder eines Containerschiffs im Roten Meer gibt es kein europäisches Produkt im vergleichbaren Segment.
A clean interface titled "Vault Dashboard," highlighting three areas of activity in a vault context with German text.
Aaaaaall the facts are checked
Es lassen sich mehrere Gründe dafür ins Feld bringen: Als Tier-1-Zulieferer der Automobilindustrie kennt Schaeffler dieses Modell der industriellen Arbeitsteilung bereits und kann es auf Robotik übertragen. Durch den Eigeneinsatz der Roboter entstehen echte Produktionsdaten, die die Aktuator-Entwicklung verbessern. Die Doppelverträge schaffen gegenseitige Abhängigkeit: Die Start-ups brauchen Schaeffler zumindest anfänglich als Abnehmer und als Lieferant. Die zwei letzten Punkte machen Schaeffler bereits zu einem attraktiven Partner in einem Markt, der noch vor der Massenproduktion steht. Die Start-ups brauchen Kunden, sie brauchen erste Referenzkunden, und sie brauchen Nutzungsdaten für die weitere Iteration der Produkte. Es gibt weitere Gründe: Schaeffler selbst lernt so ebenfalls, was dieser neue Formfaktor in der industriellen Fertigung bedeutet. Schaeffler lernt selbst, wie sie ihre Komponenten verbessern können und welche weiteren Komponenten sie entwickeln sollten. Schaeffler plant hier, das zu machen, was man in Amerika „dogfooding“ nennt: Die eigenen Produkte selbst verwenden und daraus lernen.
Claude findet heraus, dass SpaceX xAI und X übernommen hat
Nestlé caused millions of infants to die. The best estimate puts the number around 10 million. Nestlé aggressively marketed baby formula to women in poor countries. This has a number of negative effects: Formula has to be mixed with water. But in poor countries, water is often unsanitary, causing disease in infants with developing immune systems. Instructions were often printed in languages that mothers couldn’t read. Using contaminated water in formula was associated with a 27% rise in mortality. Often poor mothers would use less formula for the baby than was needed, in order to conserve it. This led to infant malnutrition. Breast milk is protective against a number of infectious diseases that can infect infants. Where infectious diseases are common, replacement with formula is harmful. When formula replaced early breastfeeding, women’s breastmilk would dry up. This led to them having to buy more formula, which they often couldn’t afford. Sometimes Nestlé would send women free formula, building dependency. Nestlé’s actions were nefarious beyond belief. One of their adverts read, “Don’t wait too long to wean your baby. If you do the little one is likely to be weak and anaemic.” They dressed over 5,000 sales representatives in nurse uniforms and sent them to hospitals to market their product to new mothers, who believed they were nurses. This would be like if Big Tobacco dressed people up as doctors to offer them free cigarettes.
Two versions of Voxel Rome are shown side by side; one is simpler with basic structures, while the other features detailed architecture.
I’ve certainly taken advantage of the new coding agents.
Microsoft AI chief gives it 18 months—for all white-collar work to be automated by AI
Vast solar panel fields stretch across the landscape, illuminated by rays of sunlight breaking through the clouds.
A user interface shows a discussion pop up window about next steps, with a chat prompt for responses.
Apple has removed more desktop Macs from its online store as the global memory shortage continues. Mac mini models with 32GB and 64GB of RAM are no longer available for purchase, nor is the M3 Ultra Mac Studio with 256GB RAM. The M3 Ultra Mac Studio is now available only in a 96GB RAM configuration, with higher-tier options eliminated. Both M3 ‌Mac Studio‌ and M4 Max ‌Mac Studio‌ models have delivery estimates of 9 to 10 weeks.
Screenshot eines Bluesky-Threads über Datacenter und Wasserverbrauch. Der Ausgangspost von geb (@geb1999.bsky.social) argumentiert, dass der Wasserverbrauch von Datacentern verglichen mit Landwirtschaft vernachlässigbar sei und Strompreise das größere Problem darstellten. Dangerous Beans (@2dangerousbeans.bsky.social) erwidert: "Farms produce food. Not a valid comparison" (219 Likes). Grumpy Rootbeer (@grumpyrootbeer.bsky.social) entgegnet: "Do they? Here in Iowa they produce ethanol and higher cancer rates" (104 Likes). gamblingpoet/lemeul/mbk (@gamblingpoet.bsky.social) ergänzt: "And Data Centers produce tech bros with stupid ideas. I'd rather have ethanol" (76 Likes). Ester Axiom (@esteraxiom.mom) fragt zurück: "you'd rather allow an industry that produces cancer than a data center?" Especially the Lies (@especiallythelies.bsky.social) antwortet darauf knapp: "I mean, kinda, yeah."
Cursor will make use of SpaceX’s massive computing resources to continue developing its own coding model Composer, launched late last year. Oskar Schulz, Anysphere’s president, said: “We think SpaceX is basically the best company in the world when it comes to building out compute. The feats they have been able to pull off are extraordinary.” Musk has been behind some of the most ambitious infrastructure projects of recent years, including the supercomputer Colossus in Memphis, which was developed in 122 days. He has staked SpaceX’s future on developing data centres in space, arguing that satellites fed by solar energy and naturally kept cool in orbit will be the cheapest way to power AI.
Iran asserts control over the Strait of Hormuz, stating it won't be fully reopened, responding to claims by Donald Trump.
NON-PROFESSIONAL & ENGAGEMENT BAIT: Anyone "classically trained" to write code pre-LLMs will take on look at the leaked Claude Code repo and see that it's a pile of unmaintainable slop. However, it is becoming increasingly clear that this doesn't actually matter. Coding is no longer a craft. No one cares how the code looks. All that matters is the utility of the program, with special exceptions made for mission critical considerations like security. Languages like TypeScript are just assembly at this point.
A UI layout shows a search bar for research, a "Council" button, and member icons, indicating collaborative AI research features.
Text in der taz "Anna-Verena Nosthoff über KI „Das Silicon Valley kommt zu sich selbst“" - Textsuche nach China hat keine Treffer
Mark Zuckerberg Is Building an AI Agent to Help Him Be CEO Meta Platforms chief uses the tool to get information faster as the company seeks to embrace artificial intelligence in all it does
Donald Trump does not think strategically. Nor does he think historically, geographically, or even rationally. He does not connect actions he takes on one day to events that occur weeks later. He does not think about how his behavior in one place will change the behavior of other people in other places. He does not consider the wider implications of his decisions. He does not take responsibility when these decisions go wrong. Instead, he acts on whim and impulse, and when he changes his mind—when he feels new whims and new impulses—he simply lies about whatever he said or did before. For the past 14 months, few foreign leaders have been able to acknowledge that someone without any strategy can actually be president of the United States. Surely, the foreign-policy analysts murmured, Trump thinks beyond the current moment. Surely, foreign statesmen whispered, he adheres to some ideology, some pattern, some plan. Words were thrown around—isolationism, imperialism—in an attempt to place Trump’s actions into a historical context. Solemn articles were written about the supposed significance of Greenland, for example, as if Trump’s interest in the Arctic island were not entirely derived from the fact that it looks very large on a Mercator projection.
Timnit Gebru piling on
the bug mag ich, aber ich bin kein großer dnb-fan. (dnb ist in eine feste formel gepresster jungle, ich hätte lieber chaotischen jungle :D)
It took Anthropic’s most advanced artificial-intelligence model about 20 minutes to find its first Firefox browser bug during an internal test of its hacking prowess. The Anthropic team submitted it, and Firefox’s developers quickly wrote back: This bug was serious. Could they get on a call? “What else do you have? Send us more,” said Brian Grinstead, an engineer with Mozilla, Firefox’s parent organization. Anthropic did. Over a two-week period in January, Claude Opus 4.6 found more high-severity bugs in Firefox than the rest of the world typically reports in two months, Mozilla said.
A dropdown menu displays options for frequency settings in German, including daily, weekly, and hourly choices and Handbuch. HANDBUCH!
Die Doku deckt alle aktuellen Features ab, einschließlich der zuletzt implementierten Text-Kategorien und Zuletzt-geöffnete-Dateien. Jede Datei enthält Architektur-Diagramme, Datenfluss-Beschreibungen, Code-Patterns und eine "Stolperstellen"-Sektion. perfekto, thx Gern geschehen! 👍
Graphs illustrate significant year-on-year growth in new websites, iOS apps, and GitHub code in the US and UK from 2022 to 2026.
A map highlighting the People's Republic of China with a visitor count of 20,519, surrounded by light blue regions.
Scatter plot showing LLMs over time with task durations to reach 50% success; recent models (GPT-5.2, Claude Opus 4.5) require multi-hour human tasks, older models minutes.
Bold headline on black background stating nuclear arms control ends this week with note about New START treaty expiration on US and Russian arsenals.
I worry a lot about the overthrow of Xi Jinping.Xi's combination of policy incompetence and political paranoia is the main thing holding back a Chinese century right now. Without him, the West might truly be cooked.
Analyse-Ergebnis: Deine Themenlandschaft Deine Kern-Themenfelder (in absteigender Frequenz): KI/LLMs & Agenten – Effizienz, Open Source vs. Closed, China-Dynamik, Infrastruktur Robotik & Embodied AI – Humanoide Roboter, synthetische Daten, Industrieanwendungen Halbleiter & Geopolitik – TSMC, EU vs. USA vs. China, strategische Abhängigkeiten Plattformökonomie – Hyperscaler, Hugging Face, vertikale Integration E-Commerce & Handel – Exchanges-Fokus, Zalando, Quick Commerce Automobil & Mobilität – EVs als Software-Plattformen, autonomes Fahren Industriepolitik – Europa-Defizite, Regulierung (AI Act, DMA) Dein analytisches Muster: Du arbeitest mit einem "Makro-zu-Mikro-Ansatz": Geopolitische Trends → Industrielle Auswirkungen → Konsequenzen für deutsche/europäische Unternehmen. Das ist dein Alleinstellungsmerkmal.
Horizontal bar chart comparing 1980 and 2024 global import shares for countries like India, Vietnam, Hong Kong, S Korea, Mexico and Poland, showing big increases.
Claude in Chrome:Computer Ein Cookie-Banner. Ich klicke auf "Einverstanden".
Headline about Trump threatening 10% Europe tariffs until a Greenland deal is reached, warning levies could rise to 25% without agreement.
Bold newspaper headline asking if Europe can afford generous state pensions, with subtext about rising elderly populations straining budgets and politics
Red book by David Marsh titled "Can Europe Survive?" with a fragmented map motif and subtitle about a continent in a fractured world.
World map highlighting reader locations with China darkest, other readers in Germany, Singapore, US, Hong Kong, South Korea, Mexico, Switzerland.
Überschrift eines Textes: Eine moderne Industriepolitik Was Deutschland jetzt anders machen muss
News article excerpt highlighting Amazon talks to invest $10bn+ in OpenAI, a deal that could value OpenAI above $500bn.
Comments thread with five users critiquing a founder’s reading list, showing avatars, names, timestamps and short sarcastic remarks about pompousness and credibility.
"Thought experiment for AI is autocomplete skeptics What’s the difference between “knowing something” vs accurately predicting what’s next?"