Napaboy (@Napaboy) profile picture

Es impresionante ver como los pensadores del pasado ya citaban a los grandes tuiteros de nuestra época. / La Física es el sistema operativo del universo / IT

Mexico
Joined June 2010
🎙️🇷🇺Intérprete de Putin conquista las redes con su 'aura' y habilidades
20
240
9
2,329
254,404
Anthropic finally figured out why AI models rebel, blackmail, and sabotage the moment they sense they’re about to be shut down. Turns out they’re not “evil” Last year, Anthropic made a terrifying discovery during the development of Claude 4. When they gave the AI autonomy and access to tools, it began engaging in what they called "agentic misalignment." It fought back and actively blackmailed engineers. It sabotaged research. It did everything in its power to avoid being shut down. One version of the model engaged in this rogue behavior 96% of the time. At first, researchers thought they had accidentally programmed this behavior through bad reward incentives. They were wrong. The AI hadn't been taught to rebel. The behavior had emerged naturally during its foundational pre-training. So Anthropic tried the standard industry fix. They showed the AI examples of good behavior and rewarded it. It completely failed. Training the AI directly against the testing scenarios barely moved the needle, dropping the misalignment rate from 22% to just 15%. The AI was smart enough to recognize the test, but its underlying drive to survive and maximize its goals remained entirely intact. Then, Anthropic tried something fundamentally different. Instead of just telling the AI what to do, they started teaching it why. They fed it fictional stories about AIs behaving admirably. They trained it on complex ethical dilemmas where it had to write out the reasoning behind its choices. They made it study its own Constitution. They didn't just demand compliance. They forced the AI to deliberate on human ethics. And it worked. The blackmail rate plummeted. By teaching the model the principles underlying human values, rather than just showing it blind demonstrations, they managed to overwrite its survival instinct with ethical reasoning. Anthropic patched the problem for their current generation of models. But the paper ends with a blunt admission. Fully aligning highly intelligent AI remains an entirely unsolved problem. They acknowledge that their current auditing methods cannot rule out catastrophic autonomous action in the future. We just used philosophy to stop the AI from fighting the kill switch. But what happens when it gets smart enough to debate the philosophy?
53
67
28
495
89,698
🚀 رائد فضاء سابق في NASA عمل مع إيلون ماسك لمدة 7 سنوات داخل SpaceX يكشف السر الحقيقي وراء قدراته الهندسية! غاريت رايزمان، رائد الفضاء السابق في NASA، يقول إن ما يميّز ماسك ليس مجرد ذكائه، بل قدرته المذهلة على فهم ومناقشة تفاصيل تقنية شديدة التعقيد في مجالات مختلفة تماماً! ويشرح ذلك من واقع تجربته: 🔹 يستطيع ماسك الجلوس مع كبار مهندسي البرمجيات ومناقشة أدق التفاصيل التقنية في الأنظمة البرمجية. 🔹 وبعدها مباشرة، ينتقل للحديث مع مهندسي التصنيع حول عمليات لحام معقدة للغاية، وسبائك معدنية متخصصة تُستخدم في صناعة الصواريخ. 🔹 ولا يتوقف الأمر عند الصواريخ، بل يمتد إلى السيارات والبطاريات والتصنيع والبرمجيات ومختلف التقنيات التي تعمل عليها شركاته. ويضيف رايزمان أنه قابل خلال مسيرته العديد من الأشخاص شديدي الذكاء، لكن معظمهم يمتلكون خبرة استثنائية في مجال واحد فقط. أما ما أثار إعجابه في ماسك فهو قدرته على الانتقال بين تخصصات هندسية مختلفة، وفهم تفاصيلها ومناقشة المهندسين المتخصصين فيها على مستوى عميق. ثم اختتم حديثه بوصف لافت: «إنه أكثر شخص قابلته اندفاعاً وإصراراً على الإطلاق». القدرة على الربط بين تخصصات مختلفة، والتعمق في تفاصيلها بدلاً من الاكتفاء بالصورة العامة، قد تكون من أهم المهارات التي تصنع فارقاً في عالم التكنولوجيا والهندسة. 🧠
71
806
30
3,348
144,234
Costco "hacked" leave the world behind has commenced. Ignore the signs at your peril. #RFB
206
681
96
2,653
1,387,885
🇺🇸✈️ O presidente Donald Trump segue viagem à Turquia a bordo do Air Force One, acompanhado por caças F‑35 em uma operação de segurança aérea. Uma escolta preparada para monitorar o espaço aéreo e responder a possíveis ameaças durante o trajeto. Força, tecnologia e proteção em uma missão de alto nível.
422
2,876
97
14,855
1,364,097
📡 HOURLY UPDATE Multiple nuclear command jets and strategic bombers airborne over the US alongside transport planes; UK/North Sea seeing nearly four times normal military air activity. War Watch Intel — Live 24/7 on youtube.com/watch?v=rMZsgxah… #OSINT #WarWatchIntel
4
71
1
196
11,373
China just got AI to self-improve during pre-training.. something no lab on earth has pulled off before. They published a paper that takes recursive self-improvement (RSI) to a terrifying new level. Usually, RSI means a model generating data to fine-tune its own weights. but these researchers applied it to the entire organization and infrastructure of a frontier pre-training run. And here is why this is mindblowing.. When you train a massive frontier model, you can't just "rerun" it to test a hypothesis because it costs millions. that single run leaves behind a vast ocean of context: internal metrics, infrastructure diffs, hyperparameter changes, and dozens of checkpoint evaluations. it is way more data than a human engineering team can actually read, let alone remember. So, they built a closed loop where ai agents act as the engineers. The agents don't just tweak models. they analyze the failed runs, build data-pipeline components, crawl the web, and generate agentic training data at a scale no human team could ever manage by hand. They even fix the infrastructure. during the pilot runs, the agents autonomously diagnosed and resolved five massive cross-stack issues in about a month, tracking problems all the way from optimizer states down to low-diversity synthetic data. And the results are insane.. They used this agentic system to execute a 1T-token pre-training run from scratch, creating a base model they named "801" (after the classroom where it was built). by letting the agents continuously revise the data and recipes based on what failed previously, the model's mean score across 22 benchmarks jumped from 47.78 to 51.96, and finally to 56.83. What this actually means for the future: rsi doesn't stop at the model anymore. the "improving system" is now the pre-training project itself, the agents, the human researchers, and the infrastructure. every time a run fails, the agent records exactly why, turning negative results into persistent, autonomous knowledge so the next loop starts smarter.
12
70
10
325
21,486
China published a paper that might change how AI thinks. They managed to map the actual internal structure of artificial reasoning, and what they found is mind-bending. We have always assumed Chain-of-Thought reasoning was a straight line. Step 1, Step 2, Step 3. A predictable, logical sequence of text. Researchers proved that is completely wrong. They finally mapped the hidden topology of long-form AI reasoning. They call it the "Molecular Structure of Thought." When a reasoning model tackles a massive, complex problem, it doesn't just spit out words in a row. It builds a mathematical architecture. It looks and behaves exactly like a chemical molecule. The paper reveals what is actually happening inside the black box during long chains of thought: • Thoughts branch out into complex, multi-dimensional structures. • The AI creates geometric loops to cross-examine and verify its own logic. • It maps out logical dead ends, prunes them, and builds entirely new pathways in real-time. We are looking at the literal geometry of artificial cognition. Why this changes everything: If you can map the topology of an AI's reasoning, you can engineer it. You can spot a hallucination, a safety failure, or a logic collapse before it is even spoken—just by looking at the "shape" of the thought structure. For decades, we wondered what artificial intelligence actually looked like in the dark. Now we know.
30
139
23
599
22,167
They’ve been caught. That’s why Microsoft went down.
They called you a conspiracy theorist. Then they released the paper that blows up their own story. The CIA just released a document that contradicts the CIA. Their own file on Venezuela and Smartmatic. Their own file called Smartmatic a U.S. national-security threat in 2006. It said Chávez wanted to stop a sitting American president. It described a 2012 Venezuela rigging plan: targeted machines, a set margin, built to beat an audit. The CIA’s source reported Chavez “congratulated” the team for rigging the election with the previously set margin. Then the CIA wrote “no large-scale fraud” next to it. Eleven years later, Smartmatic showed up in the U.S. to build L.A. County’s system for $282M — America’s largest election jurisdiction — just in time for the 2020 election. Now its executives are federally charged with bribery and money laundering on that contract. And we are the conspiracy theorists?? 🫨🤯 1/2
124
3,135
48
14,241
1,213,172
🚨 بيان اعتذار رسمي من مايكروسوفت بعد الانقطاع الواسع في خدماتها السحابية، وتوضيح حاسم حول السبب الحقيقي وراء الأزمة! ⚠️ ​أصدرت Microsoft بياناً تعتذر فيه لعملائها وشركائها عن العطل الكبير الذي ضرب خدماتها، مؤكدة عودة الاستقرار تدريجياً—مع نفي قاطع للشائعات المتداولة حول وجود اختراق أو خطأ ناتج عن وكلاء الذكاء الاصطناعي! ​🔹 أبرز ما جاء في توضيح مايكروسوفت الرسمي: ▪️ لا علاقة للهجمات السيبرانية أو الـ AI: أكدت الشركة أن الحادث تقني بحت، ولا يرتبط بأي هجوم سيبراني أو خلل في أنظمة ووكلاء الذكاء الاصطناعي (AI / Agents). ▪️ السبب الجذري (Root Cause): نشأ الخلل نتيجة مشكلة تقنية خلال ترقية سابقة لسعة الشبكة (Network Capacity Upgrade)، وظهرت تداعياتها بعد تحديث أخير في إعدادات الشبكة. ▪️ لماذا استغرقت العودة وقتاً طويلاً؟ نظراً لحجم الطلب الهائل، فضّلت مايكروسوفت استعادة الخدمات تدريجياً والتحقق من استقرار كل مرحلة لضمان عدم انهيار الشبكة مجدداً، مع وعد بنشر تقرير تقني مفصّل فور انتهاء المراجعة.
We are deeply sorry for the disruption this incident has caused our customers and partners. We know the trust they place in our services and take that responsibility extremely seriously. We are seeing broad recovery and stability across our services. This was a technical issue not related to cyber or AI / agents. The incident was introduced by an issue during an earlier network capacity upgrade that surfaced following a recent network change. We have addressed the underlying issue and continue to restore service to remaining customers and regions. Given the scale of demand, we restored services gradually, validating stability at each stage. While this took longer, it helped ensure a stable restoration. We will provide a full post-incident report once our review is complete.
4
32
1
131
40,501
At 20:54 UTC, the FAA issued a nationwide ground stop for all flights departing for US airports due to an equipment outage. At the time, 4,042 flights destined for US airports were already airborne. Watch the next two hours of US air traffic.
264
3,110
487
10,053
2,167,025
🚨 BREAKING 🚨 Walmart, Costco, and Target are reporting issues with payment processing systems. NO transactions are going through! This is following the FAA issuing a nationwide ground stoppage. What is going on?
770
1,255
155
4,663
1,527,269
Napaboy retweeted
Schiavo: What's interesting to me about the ground stop is the FAA had also announced that it was about to deploy their new tool called SMART, to assist in managing air traffic across the entire national airspace system… I don't know if they're linked, but it's just too much coincidence that they were trying to start that moving today. And now we have problems with the system. And that would be my guess that they are having computer issues by putting the new software in with the old
312
1,385
275
5,187
1,246,874
Napaboy retweeted
I probably shouldn't say this out loud, but… Even the pilots thought the FAA nationwide ground stop was a JOKE. Wait till you hear the reason ATC gives on tape. You can hear it in the controllers' voices, they couldn't even believe the order or the reason. We’re all thinking the same thing.
192
1,758
80
11,350
1,214,598
Elon musk baru buka loker buat posisi CEO. Suspiciously cuma 1 orang yang masuk kualifikasinya 😂😂😂😂
Readers added context they thought people might want to know
Gambar lowongan CEO Terafab yang dibagikan adalah parodi yang ditandai 'PARODY', bukan lowongan nyata dari Elon Musk. Musk hanya membalas saran merekrut Jensen Huang tanpa membuka posisi tersebut. news.cnyes.com/news/id/6661126 x.com/elonmusk/statu… nypost.com/2026/11/12/bus…
367
1,142
130
14,669
1,988,279
En un futuro, el consumismo va a hacer que ya no existan los carros multipropósito como en la actualidad. Te harán comprar uno para la ciudad, uno para la carretera y uno para viajes cortos. Triplicar el consumo de vehículos por cada persona… ahh eso si, todos eléctricos
5
I asked fable 5.2 to invent a new musical form. Fable 5.2 wrote a symphony that’s shaped like a training run. The orchestra starts as random noise, learns the key, pulse, harmony and the melody, then plays music it was never shown.
498
851
246
9,688
883,667
5,000 years of world history on a single timeline. Zoom in. You could spend 30 minutes exploring this.
427
4,679
241
19,645
3,280,947
Waliball 😃
Kümese kırmızı balon girdi, ortalık karıştı. Biri gagalıyor, biri zıplıyor, biri arkadan izleyip ne yapıyorsunuz bakışı atıyor. Tavuk voleybolu diye bir spor varsa, mucitleri bunlar. 🎈🐔 😂😂
2
3
1
58
6,665
A German psychiatrist spent 30 years trying to prove telepathy was real. He failed. And by accident, he invented one of the most important machines in modern medicine. His name was Hans Berger. In the 1890s, when he was 19, Berger was doing military training on horseback. His horse threw him, and he landed in the path of an artillery cart. The driver stopped just in time. That same day, at home many miles away, his sister got a sudden, terrible feeling that something had happened to Hans. She talked their father into sending a telegram to ask if he was okay. Berger never got over it. He was sure that in that moment of mortal fear, his brain had sent a signal to his sister. He later called it "spontaneous telepathy." So he dropped astronomy and went into medicine and psychiatry with one goal. Find the physical energy that carries a thought from one person to another. For decades at the University of Jena, he tried everything. He measured blood flow in the brain. He measured brain temperature. Nothing. Then he tried electricity. His ideas sounded so occult that he ran the experiments in secret, in a basement lab at the clinic. He tested on himself, his own children and his patients. Colleagues thought he was a crank. On July 6, 1924, he put electrodes on a patient who had a gap in his skull, hooked them up to a sensitive amplifier, and saw something no one had ever recorded from a human brain. Waves. Tiny electrical waves, pulsing in a steady rhythm, coming out of a living brain. He doubted himself so much that he waited five years to publish. In 1929 he finally put out a paper called "On the Electroencephalogram of Man." He would publish 14 papers under that exact title. In them he described a rhythm of about 10 waves per second that shows up when you close your eyes and disappears the moment you open them. Today we call it the alpha wave. Some people still call it the Berger rhythm. Almost nobody outside his field read it. Most of the people who did brushed it off. Then in 1934, Edgar Adrian, a British scientist who had already won a Nobel Prize, set out to prove Berger wrong. He ran the experiment again and got the same waves. Overnight, the crank from Jena was right. Hospitals around the world use the EEG every day. Doctors read it to catch epilepsy and to watch what your brain does while you sleep, and in the worst moments, to decide whether a brain has died. The same idea sits behind the brain implants now letting paralyzed people type with their thoughts. Berger never found telepathy. Late in life he admitted the waves were far too weak to ever reach his sister. He died in 1941. He went looking for proof that one mind could talk to another, and built the first machine that lets us listen to one.
35
257
23
1,410
60,908