Marc Beierschoder

@HedgieMarkets We’re overestimating the impact of a single failure. AI is already too distributed, too useful, and too embedded in workflows to disappear. The question is not if it breaks. It’s how we build around that reality.

🦔 Michael Wooldridge, professor of AI at Oxford, is warning that the race to market AI has raised the risk of a "Hindenburg moment" that could shatter global confidence in the technology. He's delivering the Royal Society's Michael Faraday prize lecture Wednesday titled "This is not the AI we were promised." His concern is that commercial pressure is pushing companies to release AI tools before their capabilities and flaws are fully understood. The scenarios he imagines include a deadly software update for self-driving cars, an AI-powered hack that grounds global airlines, or a major company collapsing because AI did something catastrophically stupid. "These are very, very plausible scenarios," he said. Wooldridge's deeper point is about the gap between what researchers expected and what we got. Many anticipated AI that computed sound, complete solutions. Instead, large language models predict the next word based on probability, leading to systems that are incredibly effective at some tasks and terrible at others, with no way to know which is which. They fail unpredictably and have no idea when they're wrong, but are designed to sound confident regardless. My Take The Hindenburg comparison is striking because that disaster didn't just kill 36 people. It killed an entire technology. Airships were dead from that point forward. Wooldridge is suggesting AI could face something similar if a high-profile failure hits the right sector at the right moment. What resonates with me is his point about unpredictable failure modes. These systems don't know when they're wrong, but they're designed to never sound uncertain. That's a dangerous combination when you're deploying them into cars, airlines, and financial systems. One quote stuck with me. He said maybe we need AIs that talk like the Star Trek computer, telling you when there's insufficient data to answer instead of confidently making something up. We got the opposite. We got systems optimized to sound authoritative whether they know anything or not. Hedgie🤗

Stephan Kohn wurde unter Angela Merkel als Beamter im Innenministerium entlassen, weil er die Corona-Maßnahmen kritisch hinterfragte. Nun war er wegen seiner kritischen Haltung als Sachverständiger von der Bundestags-Corona-Kommission eingeladen worden. Statt sich sachlich mit seinen Argumenten auseinanderzusetzen, hat die CDU versucht, ihn durch die öffentliche Bloßstellung seiner Vergangenheit als Opfer von Kindesmissbrauch zu demütigen. Solche niederträchtigen Methoden der CDU haben in einer Demokratie nichts verloren. Das ist kein Streit mehr – das ist Rufmord und ein Angriff auf den Rechtsstaat.

@McKinsey Good framework. But the real shift is not from one service category to another. It’s from selling capacity to owning outcomes. That is where the consulting model will be tested.

Agentic AI is changing how tech services create value. We’re starting to see four distinct roles take shape, each with a different set of capabilities, bets and trade-offs. The question isn’t whether to play but where to focus and how to build around it. mck.co/4mP7C7t

@pascal_bornet Interesting ranking. AI sometimes hallucinates. People sometimes disappear. My mini dachshund? Perfectly consistent… in doing exactly what she wants.

😂 After years of completely unscientific fieldwork conducted in my living room, my findings are now clear: Dogs consistently outperform. People deliver mixed results. AI is powerful, but still occasionally invents reality with confidence. That is what makes this image land so well: 👉 AI can impress you, then hallucinate. 👉 People can promise a lot, then disappear. 👉 Dogs just show up, wag their tail, and act like loyalty is a strategy. No prompt engineering, personal branding, strategic ambiguity. Just presence, consistency and results. The trust ranking is becoming awkwardly clear. What is your current trust order: AI, people, or dogs? #AI #ArtificialIntelligence #Humor #Trust #Technology #FutureOfWork #Dogs #TechCulture

Nie wieder!

@Mylovanov @ylecun Perhaps he has an end game and is just not openly communicating it? Which US president ever did?

Applebaum: Trump does not think strategically and does not have an endgame. He began the war without clear goals, never asked the American people, never spoke to Congress, and assumed it would be some kind of two-day operation that would end fast. 1/

@alliekmiller Interesting setup. But it still feels very role-centric. What we’re seeing evolve is something different: not agents as roles, but capabilities that can be dynamically composed and reused across contexts. That’s a much more scalable model than predefined “digital org charts”.

Having a 20-agent system is a million times more powerful than 100 agents working in silos. Last night, I stayed up way too late fixing and improving my digital workforce for my AI Agent Mastermind program but here are a few of the improvements I made ⬇️ - the hub and spoke model is definitely key but orchestration and documentation are the actual unlock; I gave my chief of staff (Simon) an assistant (Toby) and Toby handles the paperwork and notes and governance and reporting the daily standup - Claude kept getting confused with the word “agent” because everything is an agent so now I can refer to it as an AI CoS (Simon) with a named assistant (Toby), then under that is 6 Avengers (named after the Friends characters for ease), then under each Avenger is a set of 2-5 sidekicks (named by the task they execute), then if they need to create temporary parallelism or fanouts, they spawn “civilians” which are unnamed with no permanent file. - Moved from 5 Avengers to 6 for better task delineation (sorry Phoebe, you should have always been here! That’s my b! Might even bring in Gunther, who knows!) - Added tighter spawning rules (see image) and now allow for Simon and Toby to also spawn “civilians” - Added better post tool use to improve the documentation process - Added that Toby is keeping track of the historical team tasks to see if there are common issues to fix or promotions to make - Added an autonomy boundaries document to keep track of system-wide tool use - Changed civilians and sidekicks to use Haiku instead of Sonnet (will see if performance degrades significantly and might change back) - Improved shared external memory process because that was a dang mess - Also built out a parallel researcher with the same sidekick/civilian model to improve the speed of my research and it works like a charm ———————— 2026 agent capabilities are, in fact, insane. Will be interesting to teach hundreds of people how to do this. (You could probably take this tweet and image and ask Claude to start it for you - let me know how it goes.)

@HowToAI_ @ylecun The breakthrough is not the model. It’s when it can operate in real systems, with real data, under control.

Yann LeCun was right the entire time. And generative AI might be a dead end. For the last three years, the entire industry has been obsessed with building bigger LLMs. Trillions of parameters. Billions in compute. The theory was simple: if you make the model big enough, it will eventually understand how the world works. Yann LeCun said that was stupid. He argued that generative AI is fundamentally inefficient. When an AI predicts the next word, or generates the next pixel, it wastes massive amounts of compute on surface-level details. It memorizes patterns instead of learning the actual physics of reality. He proposed a different path: JEPA (Joint-Embedding Predictive Architecture). Instead of forcing the AI to paint the world pixel by pixel, JEPA forces it to predict abstract concepts. It predicts what happens next in a compressed "thought space." But for years, JEPA had a fatal flaw. It suffered from "representation collapse." Because the AI was allowed to simplify reality, it would cheat. It would simplify everything so much that a dog, a car, and a human all looked identical. It learned nothing. To fix it, engineers had to use insanely complex hacks, frozen encoders, and massive compute overheads. Until today. Researchers just dropped a paper called "LeWorldModel" (LeWM). They completely solved the collapse problem. They replaced the complex engineering hacks with a single, elegant mathematical regularizer. It forces the AI's internal "thoughts" into a perfect Gaussian distribution. The AI can no longer cheat. It is forced to understand the physical structure of reality to make its predictions. The results completely rewrite the economics of AI. LeWM didn't need a massive, centralized supercomputer. It has just 15 million parameters. It trains on a single, standard GPU in a few hours. Yet it plans 48x faster than massive foundation world models. It intrinsically understands physics. It instantly detects impossible events. We spent billions trying to force massive server farms to memorize the internet. Now, a tiny model running locally on a single graphics card is actually learning how the real world works.

Gehen Sie in Rente bitte. Zwischendurch mal Talkshow, bisschen plaudern. Sie sind ja sympathisch. Aber versuchen Sie bitte nicht diese Zombiepartei wieder künstlich zum Leben zu erwecken. Ihr habt zu viel Scheiss gemacht. Kann weg.

🇩🇪 in a nutshell: Ich muss 70% Steuern zahlen, mir die Ausreise genehmigen lassen, der Staat möchte meine Internetaktivitäten überwachen und wenn ich mich falsch über einen Politiker äußere, werde ich vor Gericht gezerrt. Ist das wirklich kein Sozialismus?

Wir sprechen hier von dem Wolfgang Kubicki, der während der Ampel alles mitgetragen hat? Der sich, entgegen seinen medialen Äußerungen eines Neins, höchstens der Stimme enthalten hat? Na dann viel Spaß mit der "neuen" FDP. ruhrbarone.de/kubicki-und-ha…

Wer‘s unbedingt nochmal braucht: @KubickiWo hält nie, was er verspricht. Er poltert heute freiheitlich und stimmt morgen unfreiheitlich im Bundestag. Zig mal erwiesen. Passt zur FDP.

The most dangerous lie in human history isn’t about food. It isn’t about medicine. It is about sleep. For 200,000 years, humans did not sleep 8 hours. That number was invented in 1938 by a mattress company called Simmons Beautyrest. Before that campaign, the average human slept in two shifts. Historians call it “Biphasic Sleep.” You would sleep for 4 hours, wake up for 2, then sleep for another 4. During that 2-hour window, people would pray, have s*x, write, think, and connect with their families. Some of the greatest works in human history were created in that sacred middle window. Shakespeare wrote most of his plays between 1AM and 3AM during his second wake period. Mozart composed entire symphonies in what he called “The God Hours.” Then the Industrial Revolution needed workers on a fixed schedule. You cannot run a factory on biphasic sleep. So they hired a psychologist named Dr. Nathaniel Kleitman to “prove” that 8 consecutive hours was the biological standard. He faked the studies. He was funded entirely by the mattress industry. And the medical establishment adopted his research without question because it aligned with the factory model. They turned the most creative 2 hours of human consciousness into a “sleep disorder.” They called it “Insomnia.” They medicated it. They gaslight an entire generation that 8 hours of continuous sleep was healthy. They pathologized the exact window of consciousness that produced some of the greatest art, music, and literature in human history. You are not an insomniac. You are experiencing the most natural form of human consciousness. And a mattress company convinced you it was a disease. Stop medicating your genius. Wake up at 2AM. Write the thing. The “God Hours” are calling. ✨🙌🏾💫 © Andre Gonzalves

Das Problem von Wolfgang Kubicki bei seiner Kandidatur zum FDP-Vorsitz ist nicht das Alter, sondern die fehlende Glaubwürdigkeit. Denn er hat in der Ampel jeden linksgrünen Irrsinn unterstützt, den er jetzt kritisiert, als wäre er gar nicht dabei gewesen. Wer sollte ihm glauben?

@KubickiWo Wolfgang Kubicki, 23.02.2025: „Ich kann für mich nur sagen: Sollten wir den Bundestag nicht erreichen, ist meine politische Karriere in der FDP zu Ende.“

Ich will eine Partei, die mit neuem Selbstbewusstsein die politischen Debatten in diesem Land anführt, statt ihnen hinterherzulaufen. Deutschland wartet nicht auf eine FDP, die sich in akademischen Diskussionen verzettelt, sondern auf eine, die dieses Land groß gemacht hat: klar im politischen Inhalt und ebenso klar in der Sprache. Für die Freiheit. Deshalb werde ich den Delegierten meiner Partei im Mai ein Angebot machen, bei dem sie sicher sein können: Ich werde alles tun, die Partei wieder erfolgreich zu machen. WK

@_MartinHagen Wolfgang Kubicki, 23.02.2025: „Ich kann für mich nur sagen: Sollten wir den Bundestag nicht erreichen, ist meine politische Karriere in der FDP zu Ende.“ Kein Charakter der Mann. Von Ihnen hatte ich bislang noch nicht gehört.

»Kubicki&Hagen sind die letzte Chance der FDP, aber weit mehr als ein Notnagel: Gemeinsam könnten sie die FDP nicht nur wieder in den Bundestag führen, sondern zu dem spannenden, intellektuellen liberalen Projekt machen, das die Republik dringend braucht.« ruhrbarone.de/kubicki-und-ha…

@_MartinHagen @thinkBTO @KubickiWo Kein Charakter der Mann. Wolfgang Kubicki, 23.02.2025: „Ich kann für mich nur sagen: Sollten wir den Bundestag nicht erreichen, ist meine politische Karriere in der FDP zu Ende.“

Wolfgang Kubicki kandidiert für den Bundesvorsitz - und hat mich gebeten, ihm im Fall seiner Wahl als Generalsekretär zur Seite zu stehen. Was ich sehr gerne mache. Die FDP muss zurück in die Erfolgsspur. @KubickiWo ist für diesen Job genau der Richtige! m.bild.de/politik/inland…

@BillAckman You must be one of the most stupid Americans if you believe in this bullshit. Congrats.

By far the best piece I have read on U.S. strategic options in Iran.

They should play this video for troops as they’re being sent on the ground in Iran: “Remember what you’re fighting for.”

@JonasAndrulis It’s an advantage to not belong to the first world anymore?

Europe’s hidden AI advantage: While the US sleeps, we get the premium compute. Working during global off-peak hours means our requests route to high-quality, non-quantized models by default.