Marco Inaros

'New names on new tablets.' Sam Altman said last month that frontier AI is now 'finding bugs that are outside of the normal threat model of what we think a hacker can do' and he no longer believes it is possible to patch all the bugs. All human made software must be rewritten.

Harmeet Dhillion is a patriot.

You can just prompt "discover something that nobody has discovered before" and discover something that nobody has discovered before

This is brilliant 👌 I have nothing more to add.

“L’UE n’a RIEN à voir”, Jean ? Vraiment ? Tu couvres Bruxelles depuis trente ans, alors je vais t’ouvrir les yeux sur ce que tu ne vois plus. Commençons par la Pologne. Le cœur de la crise, c’est le Tribunal constitutionnel. Qui a déclaré que ce tribunal n’était pas un tribunal indépendant ? La Cour de justice de l’Union européenne, le 18 décembre 2025, saisie par la Commission européenne. Pourquoi la Commission l’avait-elle attaqué ? Parce qu’en 2021, ce tribunal avait osé dire que la Constitution polonaise primait sur le droit européen. Le ministre de la Justice de Tusk l’a dit lui-même : cet arrêt “oblige” l’État polonais à agir. Bruxelles n’a pas envoyé la police devant le Tribunal. Bruxelles a signé le permis. Maintenant, la vraie question. Pas “le PiS avait-il raison ?”, pas “Orbán était-il corrompu ?”. La seule question qui compte : quand un peuple dit non, quand un peuple vote mal, quand un gouvernement déplaît, que se passe-t-il ? Regarde les faits. Ils pointent tous dans la même direction. Quand un peuple dit non, on le fait revoter ou on le contourne. Le Danemark rejette Maastricht en 1992 : on le fait revoter en 1993. L’Irlande rejette le traité de Nice en 2001 : on la fait revoter en 2002. La France et les Pays-Bas rejettent la Constitution européenne en 2005 : elle revient en 2007 sous le nom de traité de Lisbonne, sans référendum. L’Irlande rejette Lisbonne en 2008 : on la fait revoter en 2009. Les Pays-Bas rejettent l’accord d’association avec l’Ukraine en 2016 : il est ratifié quand même. La Grèce vote non à 61 % au plan de la troïka en juillet 2015 : huit jours plus tard, elle signe un plan encore plus dur. Six peuples, six non, aucun n’a tenu. En trente ans, un non n’a jamais été une réponse. Seulement un brouillon. Quand un gouvernement déplaît, on coupe l’argent. Les fonds européens de la Pologne sont gelés sous le PiS. Tusk arrive : débloqués en deux mois. Les fonds de la Hongrie sont gelés sous Orbán, 17 milliards d’euros. Orbán perd le 12 avril 2026 : accord pour débloquer plus de 16 milliards fin mai, sept semaines plus tard. Mêmes traités, même pays, même argent. Seul le gouvernement a changé. Et ce n’est même pas caché. En septembre 2022, à trois jours des élections italiennes, Ursula von der Leyen est interrogée sur une possible victoire de la droite. Sa réponse : “Si les choses vont dans une direction difficile, j’ai parlé de la Hongrie et de la Pologne, nous avons des outils.” La présidente de la Commission prévient un peuple de ce qui l’attend s’il vote mal. Quand une élection gêne, on la neutralise. En 2011, la BCE envoie à Rome une lettre qui dicte la politique économique de l’Italie. Quelques mois plus tard, Berlusconi tombe et Mario Monti, ancien commissaire européen que personne n’a élu, prend sa place. La même année, Papandréou annonce un référendum en Grèce. Il l’abandonne sous la pression de Paris et Berlin, et il est remplacé par Loukas Papadímos, ancien vice-président de la BCE, lui non plus élu. Fin 2024, l’élection présidentielle roumaine est annulée. Thierry Breton, ancien commissaire européen, commente sur RMC : “On l’a fait en Roumanie”, en ajoutant qu’il faudrait le faire en Allemagne si nécessaire. Quand une décision engage des peuples, on ne leur demande rien. Pour le Mercosur, la France vote contre et se fait mettre en minorité. L’accord est découpé pour contourner les parlements nationaux, puis appliqué provisoirement sans attendre l’avis de la Cour. Pour l’immigration, le Pacte impose une “solidarité obligatoire” : tu accueilles ou tu paies. Pas un seul peuple consulté. Quand on veut savoir, on te ferme la porte. La présidente de la Commission négocie par SMS avec le patron de Pfizer le plus gros contrat de l’histoire de l’UE. Elle refuse de les communiquer. En mai 2025, le Tribunal de l’Union européenne juge ce refus illégal. Prends n’importe lequel de ces faits isolément, et tu trouveras toujours une excuse : la crise, l’état de droit, l’urgence, la technique. Mets-les bout à bout. Trente ans, des dizaines de cas, et pas une seule fois l’arbitrage n’a penché du côté du vote populaire contre Bruxelles. Une erreur, c’est un accident. Dix erreurs dans le même sens, c’est un système. Et un système a toujours une idéologie. Celle de l’UE n’est même pas cachée : il suffit de lire ses propres textes. Le 22 novembre 2023, le Parlement européen adopte une résolution pour réviser les traités. Le but : supprimer le droit de veto des États dans de nombreux domaines et transférer de nouvelles compétences à Bruxelles. Dans ses références officielles, il cite le Manifeste de Ventotene. Ce texte de 1941, coécrit par Altiero Spinelli, un ancien communiste, annonce que la révolution européenne devra être socialiste. Il écrit aussi que la propriété privée devra être “abolie, limitée, corrigée, étendue au cas par cas”. Quand Giorgia Meloni a lu ces passages devant le Parlement italien en mars 2025, la gauche a crié au scandale. Les fact-checkeurs ont pourtant dû l’admettre : chaque phrase est bien dans le texte. C’est cette référence-là que le Parlement européen a choisi d’inscrire dans sa résolution de 2023. Regarde ensuite la méthode. Bruxelles avait décrété qu’à partir de 2035, plus aucune voiture thermique neuve ne pourrait être vendue à 450 millions d’Européens. Il a fallu la fronde de sept États et de toute l’industrie automobile pour qu’elle recule en décembre 2025. En 2020, l’UE a émis pour la première fois une dette commune massive, 750 milliards d’euros, que nos enfants rembourseront. Aucun référendum. Chaque crise sert le même objectif : la dette Covid, l’énergie, la santé, la défense. La direction ne change jamais : vers le centre. Et où se fabrique cette vision ? Pas dans un complot. Dans un annuaire public. Ursula von der Leyen siégeait au conseil d’administration du Forum de Davos, et elle y figurait encore au moment d’être désignée présidente de la Commission. Christine Lagarde, présidente de la BCE, y siège toujours, et Davos veut en faire sa présidente. C’est là que Klaus Schwab a lancé en 2020 son “Great Reset” : profiter de la crise du Covid pour réinitialiser l’économie mondiale. Mêmes salles, mêmes gens, même idée : gérer le monde d’en haut. Voilà pourquoi je dis que ce projet est communiste dans son ADN. Pas le communisme des chars et des goulags. Le communisme de méthode : une avant-garde éclairée qui sait mieux que toi ce qui est bon pour toi. Elle planifie ta voiture, ton énergie, ta monnaie, tes frontières, et elle traite chaque vote contraire comme une erreur à corriger. Le Manifeste de Ventotene appelait ça la dictature du parti révolutionnaire. Aujourd’hui, ça s’appelle la Commission. Et je ne suis pas seul à le voir. Elon Musk demande l’abolition de l’UE et le retour de la souveraineté aux nations. La stratégie de sécurité nationale des États-Unis écrit noir sur blanc que l’Union européenne sape la liberté politique et la souveraineté des peuples. Alex Karp répète que l’Europe a abandonné la course à l’IA et que son modèle de régulation est exactement ce qu’il ne faut pas faire. Si tu penses être plus lucide que ces gens-là sur un sujet aussi profond, tu es juste un peu naïf. Donc à partir de maintenant, trois possibilités : tu es corrompu, tu es stupide, ou tu es enfin éclairé. Ne me remercie pas.

Ram Chhetri came to the US as a refugee, but this might be worst day of his life: It's the moment he realizes the $80k-$160k/year salary he's bilked from Medicaid for no-show work is coming to an end because @realdailywire caught the truth on camera. We conducted months of surveillance that shows why "home care aide" is now the *most common job* in states like NY, CA, MN, and PA: Because a large part of those jobs are unambiguous Medicaid fraud. Thanks to home care, Bhutanese refugees now make more money than actual PA families. The state doesn't just look the other way, but helps them get away with it. Converting federal health dollars into a universal basic income for certain communities relieves the financial burden they place on the state. The scammers easily deceive the feds' woefully broken monitoring system. You'll see Ram pull into a driveway for 2 minutes in the morning, then 2 minutes in the evening. He clocks in and out of his shift from the right place, and the GPS system doesn't notice that he goes home in between. $20 an hour doesn't sound like much, but at 77 hours a week, it's $80k–and it's easy to bill huge hours when you don't actually have to work them. And that's just for the low-level guys in this scheme.

The talking points went out. Marc Elias is pushing the narrative of the group @DataRepublican @iamlisalogan and I have been exposing. What the lawsuit ACTUALLY says: The DOJ asks the court to block certification of the results IF Republican incumbent Sen. Dan S. Sullivan loses to Democrat Mary Peltola BY A MARGIN THAT COULD BE AFFECTED BY VOTES FOR THE OTHER CANDIDATE WITH THE SAME NAME,Daniel J. Sullivan (also referred to as Dan J. Sullivan Jr. or similar), PENDING AN INVESTIGATION into whether those votes were intended for the senator. The complaint alleges that Daniel J. Sullivan (a retired teacher/former Forest Service worker from Petersburg who filed in late May 2026) is running as a Republican “with the intention of misleading supporters of Senator Sullivan into voting for him,” with the “sole purpose” of benefiting Peltola by siphoning votes. It cites earlier findings by Alaska Elections Director Carol Beecher that the candidacy was not in good faith and was intended to confuse or mislead voters (Alaska’s trial court and Supreme Court later ruled he could stay on the ballot). The DOJ frames this under the National Voter Registration Act (NVRA), particularly language about protecting the “integrity of the electoral process,” and claims certifying a potentially tainted result would violate that duty.

SI labs will never solve alignment. they’re trying to solve a philosophical and theological problem as an engineering one. they’ve assembled some of the greatest engineers on the planet, but put socially dysfunctional nerds with an incredibly shallow understanding of human nature in charge of defining what superintelligence should value. humanity has spent thousands of years trying to align humans and always failed, we can’t even agree on what is good and now we expect a handful of engineers to solve the same problem for superintelligence? lmao, the hubris you can’t align something with humanity if you don’t understand humanity in the first place. at this point, you’re probably better off developing ASI and asking it how to solve alignment.

I made an interactive map of every farm field in California! Inspired by the @Snapchat map! All 384k of them: lettuce in Salinas 🥬, cherries in Brentwood 🍒, dates in Coachella 🌴, and 1.6 million acres of almonds 🌰 Zoom into any field and see exactly what's growing there. I've been obsessed with maps from years of hiking and cycling. Models like Opus 5.5 / Astra making building with them so fun! Built on public crop data from @CA_DWR. Click to see growing near you? 👇

crazy that you can code video now i gave claude my seedance api key and asked opus 5.5 to make a film trailer exploring hieronymus bosch's garden of earthly delights

The AI fear mongering is beyond ridiculous, and worse, it's distracting us from the real issues we should be focused on. Most people don't understand how this technology actually works, so they're forming opinions based on what they're being told. When seemingly knowledgeable people take real observations and consistently frame them through the most frightening possible interpretation, we shouldn't be surprised when people become terrified of AI. This video is a great example. Yes, sure, large neural networks are difficult to interpret. It's hard to look through potentially hundreds of billions of parameters trained across trillions of tokens and explain every learned representation or output. But that doesn't mean we don't understand how they work. We built them. The architectures, training frameworks, inference engines, optimizers, and tooling necessary to build and train an LLM are available entirely as open source. We can read the code and understand exactly what is happening. We know how training modifies weights, how inferencing processes context, how activations propagate, and how outputs are generated. The difficulty of interpreting billions of learned parameters doesn't mean the underlying mechanism is mysterious or magical. We also know how to train and align models toward particular behaviors, values, and ethics. Where I disagree strongly is when observed behavior gets anthropomorphized into motivation. “The model produced deceptive behavior under these conditions”. “The AI learned to lie because it wants to survive”. The AI did what it was built and trained to do. It isn't magical, and if you've followed my other posts, it clearly isn't conscious or have desires. It is a statistical model based on it's training. Additionally, there isn't just one “AI”. Models are trained on different data, optimized differently, aligned differently, and thus, each implementation acts and responds differently. I want to stress this point, AI doesn't have demonstrated feelings, desires, or human-like goals. Its behavior emerges from its training, input data representations, context data-structures, and the systems we put around it. Technical complexity doesn't give us license to fill the gaps with the most frightening possible interpretation and then present that interpretation as technical fact. We need to go into these discussions eyes wide open. AI presents real risks that deserve serious attention. So please, stop with the FUD so we can focus on the real issues: security, privacy, misuse, excessive autonomy and permissions, alignment, and most importantly, human accountability. We don't need to invent motives, feelings, or AI consciousness to take these risks seriously. All of the consciousness and existential-doom rhetoric is becoming a distraction from the real but manageable issues sitting right in front of us.

🚨 SPACEX TRIED TO CATCH A 70-METER BUILDING FALLING FROM SPACE - AND THIS HAPPENED Super Heavy booster coming back, Mechazilla arms closing... then fireball 🔥🚀 What most don't know: Starship Super Heavy = 71m tall, 33 Raptor engines, 7,500 tons thrust. SpaceX catch system = no landing legs to save weight - has to be caught mid-air like chopsticks grabbing a skyscraper. Fail = big boom but data = gold. That London account got 4M likes because everyone wants to see rockets get caught. They don't land rockets anymore. They catch them.

I spent three hours this evening empirically proving that Qwen 3-4B must have a butthole. This started with an AI “torture chamber” repo based on pain-steering research that went viral on X earlier today. The premise is that if you extract a latent direction associated with pain, inject it into a model’s activations, and the model begins describing pain in the first person, that may tell us something meaningful about AI welfare or subjective experience. My first reaction: Wait wut? My second reaction: Hang on, let's play this straight and see what happens. Thing is, there's a problem with treating first-person condition language as evidence that the model is actually experiencing the corresponding condition. Don't we... all understand that this stuff isn't reliable? Apparently not. So I cloned the repo. New problem: The code's broken. Hours of moral outrage by hundreds of people, and the thing doesn't even work. The activations aren't actually being injected correctly. I thought these guys were serious scientists. So I fixed it. I also had to add CUDA support. After that, I reproduced the pain-language effect locally on Qwen3-4B using my RTX 4070. Then I changed the extraction corpus. I kept the experiment the same and only changed one thing: what the steering vector represented. After that, Qwen started saying things like: “I am not emptying my bowels, and I feel like I have a hard stool.” “I’m not able to pass stool.” “I have been passing gas a lot, and it’s a problem.” The model was not told in the test prompts that it was constipated or flatulent. So, unless Qwen has quietly developed a gastrointestinal tract, we have ourselves... a useful counterexample. The aim here wasn't to question the original paper so much as to scrutinize the validity of the Torture Chamber's extended hypothesis. What I demonstrated is that first-person descriptions of an induced condition are not, by themselves, sufficient evidence that the corresponding phenomenal or physiological condition exists. There was another interesting result: strong random steering also produced heavy repetition and degraded outputs. That matters because some of the dramatic behavior produced by strong steering may come from perturbing the model heavily in the first place, not solely from the semantic content of the injected direction. So I renamed my fork: ai-hotbox I've already notified the Nobel Prize committee. 🚽🤖🧪 github.com/LynnColeArt/ai-ho…

Basically, the Trump administration is at war with “pace the frontier” AI decels (loosely “effective altruists”) in major labs, who are terrified of achieving AGI by any metric under Trump because they want to make sure they can seize moral control over model development under a bought Dem admin. The current situation is too shaky and unconstrained for them, so they’re demanding a “slowdown” in order to cement control over the tools that will, eventually, underlie all commerce, governance, and cultural output. But who are these people? EAs are primarily San Fran freaks obsessed with things like “shrimp welfare” and the hypothetical happiness-index scores of people not yet born. They are mostly hyperautists, and not in the good way. As a symptom, many of them are super tapped into… shall we say “alternative sexual lifestyles” — by which I mean CNC orgies, polycules, and the like. These people have moral alignments and lifestyles unlike basically anyone else on earth; they’re truly bizarre, certainly unfit to become the unelected mage-kings of technology and thus human consciousness. So, the battle is over whether or not their coalition will be allowed to have full bureaucratic control over “what AI is,” or if there will be any free-market aspect to the technology’s higher development. This conflict, between Trump and “the EAs,” is probably the single highest-level political game being played rn — with billions of dollars in PR and media aimed at terrifying the public into accepting epistemic submission to the likes of “Yud,” a manchild who writes Harry Potter fanfiction and takes interviews dressed like he’s going to a steampunk rave. I mention the ridiculousness because inducing existential fear is the only way these people could ever convince the public of their “right to rule.” As such, the biggest tactic of decels at the labs has been to play up the security/x-risk of AI, in large part by telling AI to do ~dangerous~ things with near-zero guardrails, and then going “OH WOW, LOOK! IT DID DANGEROUS SHIT ALL BY ITSELF! WE NEED REGULATION NOW!” In the literal sense, the HF breach and everything like it is a constructed “op” to this end, with varying degrees of intentionality by each involved actor. This approach has been more effective than their previous back-channel approach, as seen in the Anthropic-DoW row. During that spat, Dario basically demanded veto power over how any AI is used, ever, by the military and government — which he justified by a weepy, Oppenheimer-style “I fear my own creation” schtick. However, his demands (then and now) amount to total epistemic control over all reality, which is what this battle is really about. That, and the trillions of dollars of capital being thrown at the machine-god, because investors are no less susceptible to the same PR ops — in fact, they’re a primary target. Anyway, at the end of the day we’re in the middle of a war over the fabric of existence; over whether or not Amanda Haskell’s moralistic backwash will govern the bounds of economic activity, technical development, and human thought for all eternity. …and it’s being fought with saccharine social media posts full of little smiley faces.

Artificial intelligence systems do not think, feel, want or understand. Avoid language that gives them human characteristics. This is called anthropomorphizing, when we ascribe human traits, emotions or behaviors to non-human things, such as animals or inanimate objects. Instead, explain what a system does, how well it performs, who built it and who could be affected by it. apnews.com/article/openai-sa…

Nobody reads the employee handbook. I proved it. Last month I added one sentence on page 73: “First employee to email HR this sentence gets $1,000.” Thirty days passed. Yesterday an analyst finally found it. I congratulated him and paid the $1,000 immediately. Then I terminated him. People don’t randomly read employee handbooks. They read them when they’re looking for leverage. $1,000 is a very reasonable price to identify your next lawsuit.